AlphaContext: An Evolutionary Tree-based Psychometric Context Generator for Creativity Assessment Yixuan Wang1 * Yue Huang1 * Hong Qian1,2† Wenkai Wang1 Zhi Liu1,2 Zhongjing Huang1 1
Yunzhao Wei1 Aimin Zhou1,2
Yifei Ding1 Jiajun Guo1
East China Normal University, Shanghai, China Shanghai Innovation Institute, Shanghai, China
2
{yxwang, yhuang}@stu.ecnu.edu.cn, [email protected] Abstract
arXiv:2604.18398v1 [cs.CL] 20 Apr 2026
Creativity has become a core competence in the era of LLMs and human–AI collaboration, underpinning innovation in real-world problem solving. Crucially, the systematic improvement of creativity necessitates scientifically valid assessment instruments. Psychometric research recognizes context-based assessment as an effective way to measure creative thinking. However, high-quality expert-designed contexts remain scarce. Existing LLM-based generators often struggle with insufficient assessment cues, weak narrative coherence, limited stylistic diversity, and poor support for creative thinking. To address these challenges, we propose AlphaContext, an evolutionary tree-based psychometric context generator for creativity assessment. First, the HyperTree Outline Planner formalizes expert-designed outlining as a rule-guided hypertree and performs top-down hierarchical planning. The MCTS-based Context Generator fills the outline via MCTS to balance global structure and local quality. Then, the Evolutionary Context Optimizer evolves contexts with MAP-Elites by repeatedly updating niche elites to jointly improve diversity and quality. Finally, the Assessment-Guided Evolution Refiner simulates virtual participants with diverse styles and recycles weak contexts for further evolution. Experiments show that AlphaContext yields an average improvement of 8% over competitive methods across 6 quality metrics.
1
Introduction
Creativity is typically defined as the ability to generate novel and appropriate ideas (Runco and Jaeger, 2012), and it is a crucial skill that drives social innovation and scientific discovery (Sternberg and Lubart, 1999). As AI increasingly takes over routine tasks, creativity is becoming an even more important driver of original contributions and transformative breakthroughs (Lee, 2022). * Equal Contribution †
Corresponding Author
Context Design Process
Context: AI Partner
companionship human–AI ethics...
AI Partner
Design Principles
[Arts & Aesthetics][Basic Needs][Business & Commerce][Comm unication][Culture & Religion][.....]
[Anchor] A gentle melody streamed from the smart speaker by Emma’s bed, while the ambient lighting gradually shifted to a fresh, pale blue. This was the “Morning Wake-Up Service” provided by her AI partner, Q-Bot. It could accurately detect Emma’s sleep stages and begin her day uniquely with the most suitable Design combination of sound and light. At its the core, Q-Bot was an intelligent system Context connected to the cloud—its human-like form merely served as an interface for real-world interaction [……]
[Anchor][Scene Setting] [Characters & Interaction] [Conflict & Challenge][Open Task]
[Open Task] Please identify the potential challenges arising from the widespread adoption of “AI Partner”, listing as many as you can. Complete the Assessment
Creative Challenge Cue Association
Answer
human-like form
detect sleep stages
[Ethics & Morality]
felt lonely
[Psychological Health]
Creativity Context Assessment
Answer 1
[Physical Health]
a fresh,pale blue
Participant1 Answer 2
[Arts & Aesthetics]
Answer 1 Participant2 Answer 2
Figure 1: Workflow of creativity context assessment. Experts first design an “AI Partner” psychometric context containing implicit challenge cues. Participants then complete an open-ended task to identify potential challenges based on the context in their responses.
Given this growing importance, scientifically assessing creativity has become a key issue in psychometrics and intelligent education (Jr. et al., 2024; Wang et al., 2025b; Guo and Woulfin, 2016). In creativity assessment, context-based paradigms have been widely adopted (Barbot et al., 2019b). Sternberg’s triarchic theory of intelligence emphasizes that creative thinking arises when faced with novel situations (Sternberg, 1984). Therefore, futureoriented contexts, due to their inherent high uncertainty and broad imaginative space, are considered ideal stimuli for eliciting creativity thinking (Koh and Leung, 2019), as shown in Figure 1. The Future Problem Solving Program (FPSP) (Torrance et al., 1976) provides authoritative evidence that well-designed future contexts can reliably elicit creative thinking (Crabbe, 1982). Therefore, generating high-quality contexts is essential for developing valid and reliable creativity assessments
for humans. However, current practice faces a significant bottleneck in productivity. High-quality creativity assessment contexts still rely on expert craftsmanship. In recent years, rapid advances in LLMs have substantially improved the generation of stories and dialogues (Li et al., 2025a; Feng et al., 2025; Wu et al., 2025a), making the automated construction of psychometric contexts for the assessment of creativity increasingly plausible. Although much prior work has examined the creativity of LLMs, there has been far less research on whether LLMs can generate valid creativity assessment contexts for humans. However, psychometric contexts differ from general narratives, and both directly influencing LLMs and reusing story-generation frameworks still face two key challenges. The first key challenge is to simultaneously enforce implicit assessment cues and global narrative coherence. Psychometric cues for creative thinking are often embedded implicitly in textual details. Existing methods struggle to precisely control the alignment between cues and themes (Bai et al., 2025), thus failing to satisfy the fine-grained constraints required for psychometric content design and narrative structuring. The second key challenge is to improve diversity while ensuring both context quality and measurement validity at limited cost. For a given theme, future problem contexts require diverse types and styles to adapt to different assessment populations (Novikov et al., 2025), yet increasing diversity typically raises generation and refinement costs. Moreover, creativity assessment contexts require reliable quality and elicitation validity. Traditional expert workflows as shown in Figure 1, rely on expensive human studies and iterative rework (Wu et al., 2025b), while current methods lack efficient validation and optimization loops. Our work has implications for employing LLMs to automatically generate valid and reliable creativity assessment contexts for humans and AI. To address these two challenges, this paper proposes AlphaContext, an evolutionary tree-based psychometric context generator for creativity assessment. To tackle the first challenge, the HyperTree Outline Planner formalizes context outlining as a rule-guided hypertree, mapping expert reasoning into a searchable outline space. The MCTSbased Context Generator then performs Monte Carlo Tree Search (MCTS) generation under the outline, balancing global structural coherence and local semantic quality to produce seed contexts. To
handle the second challenge, the Evolutionary Context Optimizer conducts evolutionary search with MAP-Elites in a task-specific behavioral space, iteratively expanding stylistic diversity via nichewise elite updates. Finally, the Assessment-Guided Evolution Refiner simulates virtual participant responses and iteratively refines weak contexts to better elicit creative thinking. Experimental results show that AlphaContext substantially outperforms baselines across multiple evaluation metrics.
2
Related Work
2.1
Language Model based Creativity Assessment
The NLP community has shown growing interest in psychometrically grounded creativity assessment. Luchini et al. (Luchini et al., 2025) finetuned RoBERTa and GPT-2 to automatically score creativity responses. Since assessment contexts are expert-designed and costly to scale, CPIG investigates using LLMs to automatically generate items for a classic free-response creativity test, examining whether LLMs can generate valid creativity assessments for humans. However, CPIG targets short items and does not address long-form Future Problem contexts, which require discourse-level coherence and implicit assessment cues. In parallel, other works evaluate the creativity of LLMs (Si et al., 2025; Fang et al., 2025). LiveIdeaBench (Ruan et al., 2024) uses singlekeyword prompts to assess scientific creative thinking. AidanBench (Mclaughlin et al., 2024) measures novelty, non-redundancy, and coherence under open-ended creative questions. Nevertheless, these benchmarks rely on keyword triggers or expert-designed contexts. Moreover, the scarcity of high-quality long-form contexts limits both evaluation protocols and model improvement. Overall, the key bottleneck is generating highquality long-form psychometric contexts that enable valid creativity assessment for humans and also support more comparable evaluation of LLM creativity. To address this gap, we propose AlphaContext, an evolutionary tree-based psychometric context generator for creativity assessment. 2.2
LLM-based Story Generation
LLMs have significantly advanced automated narrative generation (Bai et al., 2025; Wu et al., 2025a; Lee et al., 2025), enabling the generation of coherent long-form stories. DOC (Yang et al., 2023)
(a) Model Input Context Query
(b) Tree-based Context Generation
(c) Context Evolution and Refinement
HyperTree Outline Planner
Evolutionary Context Optimizer
Assessment-Guided Evolution Refiner
Operators
Simulated Participants
LLM
LLM
HT-Select HT-Expand
Title
MCTS-based Context Generator
HT-Decide HT-Construct
MCTS-Select MCTS-Expand MCTS-Evaluate MCTS-Backpropagate
Insertion Deletion Replacement
LLM
Talkative Normal Quiet
LLM
Theme … Answers
Mutation Design a future-oriented creativity problem context titled “AI Partner” about human–AI companionship and autonomy, focusing on ethics, emotional reliance, privacy, and governance of pervasive personal AI assistants.
Evaluation
… Outline
Seed Context
Evolve
LLM Scorer
LLM
Creativity Scores
Archive
Elite Update
Output Context
Figure 2: The procedure of the proposed AlphaContext. (a) Given a context query Q, (b) the HyperTree Outline Planner and MCTS-based Context Generator generate seed contexts, and (c) the Evolutionary Context Optimizer and Assessment-guided Evolution Refiner improve diversity and quality, yielding assessment-ready contexts.
adopts the outline-first strategy and then expands the outline into detailed text. STORYTELLER (Li et al., 2025a) introduces a plot node mechanism based on the subject-verb-object (SVO) structure and a dynamic interaction module, further improving narrative coherence and logical consistency. However, most existing story generation methods typically focus on entertainment and fluency, often failing to satisfy the quality and validity requirements of psychometric assessment contexts. Although SS-GEN (Feng et al., 2025) explores generating psychological social stories for autism interventions, this setting differs fundamentally from creativity assessment settings. Creativity assessment contexts must maintain coherent long-form narratives while implicitly placing assessment cues that elicit creative thinking and support psychometric validity. Consequently, generating high-quality long-form psychometric contexts for creativity assessment remains an open challenge.
3
Preliminaries
Creativity Context Generation. Given a context design query Q that specifies the title and theme, together with a pre-trained LLM πθ , our goal is to construct an assessment-ready archive |A| A = {Ck }k=1 of contexts for creativity assessment. Each context is represented as C = (T, O), where O is a structured outline, and T is the resulting context text guided by O. We adopt a plangenerate-evolve pipeline. The planner produces O = πθ (Φ(Q)) under predefined instructions Φ. The generator realizes T guided by the outline O.
The evolve stage iteratively refines and diversifies candidates by updating A. HyperTree Structure. In conventional tree-based planning, each edge links a parent node to a single child node. By contrast, a HyperTree introduces directed hyperedges, where a parent node connects to a set of child nodes via one edge, enabling hierarchical divide-and-conquer by jointly organizing discourse structure and assessment-cue placement for outline planning. Formally, we define a HyperTree as H = (N, Q, R), where Q denotes the query, N is the node set, and R is a set of expansion rules. Given Q, the HyperTree is generated hierarchically according to the rule set R. Compared with ordinary trees, this structure better aligns with expert practices in creativity context design.
4
The Proposed AlphaContext
Overview. As shown in Figure 2, AlphaContext comprises four modules: the HyperTree Outline Planner, the MCTS-based Context Generator, the Evolutionary Context Optimizer and the Assessment-Guided Evolution Refiner. Given a query Q, the HyperTree Outline Planner places expert outlining as a rule-guided hypertree over a library R. The MCTS-based Context Generator then performs a sentence-level search to fill in the outline. To cover the multi-solution space under the same theme, the Evolutionary Context Optimizer applies MAP-Elites to explore diverse styles in a task-specific behavior space while improving within-niche quality. Finally, the AssessmentGuided Evolution Refiner simulates participant re-
sponses and feeds weak contexts back for further evolution.
rative consistency with the input title and theme. More details can be found in Appendix G.
4.1
4.2
HyperTree Outline Planner
Experts plan contexts holistically and refine them hierarchically, motivating a HyperTree representation. We propose the HyperTree Outline Planner to cast outline design as a HyperTree search, where directed hyperedges support hierarchical divideand-conquer over structure and cue placement. Formally, we define a HyperTree as H = (N, Q, R). Given Q, HyperTree H is generated hierarchically under R. Each node n ∈ N corresponds to a structural unit, and each rule r ∈ R expands a parent node into a set of child nodes through r : np 7→ nc , where np is the parent node and nc denotes the corresponding child nodes. The HyperTree (HT) planner proceeds in four phases: HT-Select, HTExpand, HT-Construct, and HT-Decide. HT-Select. Given the current hypertree H, its distinct branches are mapped onto a set of hyperchains {L1 , . . . , Lg }, where g denotes the number of hyperchains. To control the search scale, an LLM is used to evaluate and prune candidate hyperchains. Each candidate hyperchain is scored by the LLM and the optimal hyperchains L∗ are selected. The divisible nodes are then identified under the rule set R. For each selected hyperchain, we choose its most promising divisible leaf node for expansion using an LLM selector. Thus, this phase consists of two steps: selecting the hyperchains L∗ and selecting the divisible leaf node n∗i in each chosen hyperchain L∗i ∈ L∗ . HT-Expand. In this phase, given the selected node n∗i in each hyperchain L∗i , its applicable expansion rules are retrieved as R(n∗i ) = { r ∈ R | r : n∗i 7→ nc }. For each rule r ∈ R(n∗i ), candidate child groups nc are generated. Each group is treated as a single branch and is appended to L∗i as the expansion outcome of n∗i . HT-Construct. The planner iterates Select and Expand step by step, growing H from the root node across depth levels over a set of selected hyperchains. Construction stops when no divisible nodes remain or the iteration limit is reached, yielding a hypertree that compactly stores multiple candidate hyperchains. HT-Decide. After constructing H, the LLM globally evaluates the candidate hyperchains and decides the optimal hyperchain as the final outline O. This decision jointly considers structural validity for creativity assessment context design and nar-
MCTS-based Context Generator
In the generation stage, we cast creativity context writing as a sentence-level decision process guided by an outline O. Given an input prompt x and an outline O, we build a separate search tree for each discourse section. A context is represented as C = (T, O), where T = (t1 , . . . , tp ) is the generated sentence sequence. A node at depth p is sp = tp , N (sp ), V (sp ), O , where tp is the current text, N (sp ) is the visit count, and V (sp ) is the estimated value. An LLM policy πθ proposes the next candidate sentences, and an LLM evaluator provides quality feedback. The generator follows the standard MCTS loop: MCTS-Select, MCTS-Expand, MCTS-Evaluate, and MCTS-Backpropagate. MCTS-Select. Starting from the root s0 , the search recursively selects the child with the highest exploration potential according to the UCT score. Specifically, the UCT score ofqthe node sp is defined as
(q) UCT(sp ) = V (sp ) + c lnNN (sp ) . Here, V (sp ) denotes the value score of sp , N (sp ) is its visit count, and N (q) is the visit count of its parent node q. c is a hyper-parameter that balances exploitation (V (sp )) and exploration (the second term). MCTS-Expand. Given the selected node sp , we expand it by sampling U candidate next sen(u) tences from the policy model πθ : tp+1 ∼ πθ (· | x, t1:p , O), u = 1, . . . , U . Here, t1:p denotes the previously generated sentences, enabling parallel exploration of diverse narrative realizations and cue instantiations within O. MCTS-Evaluate. We evaluate each expanded node to assign its node value V (sp+1 ). For longform creativity context generation, we adopt a dual-horizon valuation mechanism at the evaluation phase to balance reliability and computational cost. Given an expanded child sp+1 , we first apply multi-aspect immediate scoring with an evaluator:
Vimm (sp+1 ) = S̄(sp+1 ) 1 − Sha (sp+1 ) .
(1)
Here, S̄(sp+1 ) is a weighted average of cue alignment Ssc , imageryP vividness Sim , and discourse coherence Sco with i ωi = 1. Sha (sp+1 ) measures hallucination risk. To mitigate myopic decisions, when Vimm (sp+1 ) < τ , we sample a short continuation and re-evaluate the concatenated fragment to obtain a more stable value estimate, a lightweight
look-ahead for this node. This look-ahead is triggered for low-scoring nodes to reduce myopic errors, while high-confidence nodes directly use the immediate evaluation to save sampling budget. For more details, please refer to our Appendix F. MCTS-Backpropagate. The obtained evaluation score re is propagated back along the simulated path to all ancestor nodes sj (0 ≤ j ≤ p), updating the visit counts and value estimates: Nnew (sj ) = Nold (sj ) + 1 , Vnew (sj ) =
Vold (sj ) Nold (sj ) + re . Nnew (sj )
(2) (3)
After multiple simulations, the tree concentrates on trajectories that better satisfy the outline, improve coherence, and reduce hallucination risk. We then extract the highest-value root-to-leaf path as a seed context to initialize the evolutionary module. 4.3
Evolutionary Context Optimizer
We introduce a MAP-Elites Evolutionary Context Optimizer initialized with the MCTS seed context. It maintains an elite archive in a style-oriented behavior space, expanding coverage and improving within-niche quality. We next describe the archive, mutation, evaluation, and update rules. Diversity Archive. To characterize stylistic variations in creativity assessment contexts, we map each candidate context C into the behavior space B with a descriptor function b(·): b(C) = ϕ1 (C), ϕ2 (C), ϕ3 (C) ∈ [0, 1]3 . (4) Here, ϕ1 captures proximity scope, measuring the extent to which a context is framed from personal daily-life settings to broader public issues. ϕ2 captures knowledge density, reflecting how strongly the narrative is grounded in objective evidence such as data, mechanisms, and causal explanations. ϕ3 captures viewpoint diversity, indicating the breadth of stakeholders involved and the need for multiperspective integration. We uniformly discretize [0, 1]3 to form a 3D grid archive, where each cell defines a behavioral niche and stores the current elite context with the highest fitness. Mutation. In natural language space, we implement mutation as a conditional LLM editing policy πθ that edits a parent elite context Cp . At each iteration, we apply an operator set Ω = {I NSERTION, D ELETION, R EPLACEMENT} to revise key paragraphs and cue-bearing units,
introducing stylistic variation while maintaining assessment-critical content. Evaluation. Given a mutated candidate C, we perform both behavior feature evaluation and quality evaluation. Feature evaluation computes b(C) to determine the candidate’s niche assignment. Quality evaluation is produced by an LLM-based scorer in terms of narrative coherence Scoh , topical relevance Srel , and engagement Seng , and we de-
fine fitness F (C) = Avg Scoh (C) + Srel (C) + Seng (C) . The scorer outputs three normalized quality scores for each candidate context, and the fitness is defined as their uniform average to avoid introducing extra hyperparameters. Elite Update. After evaluation, we assign context C to the grid niche indexed by b(C). If the niche is empty, C is inserted as the initial elite. Otherwise, let C ∗ denote the current elite in the niche; we replace C ∗ only when F (C) > F (C ∗ ). 4.4
Assessment-Guided Evolution Refiner
To ensure that generated creativity contexts reliably elicit measurable creative thinking, we propose an Assessment-Guided Evolution Refiner. Concretely, we instantiate an LLM-based participant simulator with a temperature set to 0 to suppress sampling randomness and stabilize response generation, making contexts comparable in a consistent simulation setting. To improve realism and interpretability, we model response styles as explicit profiles grounded in common psychometric response patterns and enforce each profile via role-conditioned prompting. We consider three profiles—talkative, normal, and quiet. Given a candidate context C, the simulator generates a set of responses {Ym }M m=1 . The refiner then scores each response using a creativity scorer fcre (·). We define the assessment effectiveness of a context as the average creativity score across 1 PM styles Ψ(C) = M f m=1 cre (Ym ). If Ψ(C) exceeds an expert-specified threshold, we treat C as an assessment-ready context. Otherwise, we route C back to the Evolutionary Context Optimizer for further iterative optimization.
5
Experiments
This section first describes the CreaTE dataset and details the evaluation metrics used. We then conduct extensive experiments to answer the following research questions. The codes and data are available at https://github.com/yxwang19/ AlphaContext.
Table 1: Performance comparison across different methods on the CreaTE dataset. AlphaContext achieves the best results across all seven perspectives. All metrics are presented as positive percentages, where higher values indicate better performance. For each metric, the best-performing model is highlighted in bold and the second is underlined. Coherence (↑)%
Relevance (↑)%
Engagement (↑)%
Significance (↑)%
Concreteness (↑)%
Uncertainty (↑)%
Diverse Verbs (↑)%
DeepSeek-V3.1 Qwen3-235B-A22B Llama3.3-70B-Instruct LongWriter-llama3.1-8b LongWriter-glm4-9b GPT-5.1 Gemini-3.0-Pro-Preview DOC-v2 CRITICS SS-GEN
50.00 41.50 27.83 26.60 32.27 70.44 72.54 49.14 51.11 60.22
50.00 47.91 26.97 27.46 31.40 70.20 75.37 61.33 61.95 69.69
50.00 45.07 33.99 28.63 31.38 65.39 62.56 61.82 61.21 56.40
50.00 33.87 27.46 23.40 36.19 50.37 48.40 34.98 37.81 60.10
50.00 46.55 30.54 33.62 32.98 71.80 64.16 51.11 54.43 51.85
50.00 53.33 20.07 25.99 31.32 68.60 63.30 43.10 42.12 53.57
94.33 92.09 90.39 91.18 88.69 92.88 91.81 92.82 92.31 90.24
AlphaContext
81.28
79.06
79.93
71.06
75.49
80.30
96.06
Methods
Q1: How does AlphaContext compare to existing methods in generating high-quality creativity contexts across multiple evaluation dimensions? Q2: To what extent do the core components contribute to the performance of AlphaContext? Q3: How does AlphaContext compare to other methods in terms of textual similarity with expertdesigned contexts? Q4: Does the LLM-based judge align with human preferences to support reliable evaluation? Q5: Does AlphaContext remain effective for creativity assessment in real-world human studies? Q6: How does the correlation of AlphaContext with expert-designed assessments compare to that of the strong LLM baseline? Q7: What is the computational cost of AlphaContext in terms of generation time and token consumption? 5.1
Experimental Setup
Dataset. We evaluate AlphaContext on CreaTE. Since general story-generation datasets prioritize narrative completeness and style rather than assessment alignment, we construct CreaTE: 203 expert-curated title–theme inputs balancing evaluation cost and domain coverage. Three creativitypsychology experts write each entry and conduct iterative cross-checks. An entry is included only after consensus validation and revision, and all inputs are further screened to remove sensitive information. Details are provided in Appendix A. Baselines. To compare AlphaContext with both strong LLMs and generation frameworks, we consider baselines from three categories. The prompt template is provided in Appendix I. (i) General-purpose LLMs. We consider
DeepSeek-V3.1, Qwen3-235B-A22B, Llama3.370B-Instruct, GPT-5.1, and Gemini-3.0-ProPreview as competitive instruction-following models with strong general reasoning and generation capabilities. (ii) Long-form specialized LLMs. LongWriter-llama3.1-8b and LongWriter-glm49b (Bai et al., 2025) are included as specialized baselines for long-form writing. (iii) Structured generation frameworks. DOC-v2 (Yang et al., 2023) combines hierarchical outlining with an adherence controller. CRITICS (Bae and Kim, 2024) performs critic-guided iterative refinement. SSGEN (Feng et al., 2025) applies constraint-driven hierarchical prompting (STARSOW) for structured story generation. We do not compare with CPIG since it generates short test items and is not designed for long-form context generation. Evaluation Metrics. We evaluate each generated creativity context along 7 dimensions. Coherence measures narrative consistency, Relevance measures theme alignment, and Engagement measures how motivating the context is for participants. Our evaluation framework is theoretically grounded in Amabile’s Componential Model of Creativity (Amabile, 1983, 2018), which emphasizes task motivation, domain-relevant grounding, and creativity-related processes as core foundations of creative performance. We further evaluate Significance (Okuda et al., 1991; Mumford et al., 2018) to capture real-world relevance and intrinsic task motivation, Concreteness (Guegan et al., 2017) to reflect situational specificity that supports feasible ideation, and Uncertainty (Beghetto and Jaeger, 2022; Beghetto, 2021) to quantify open-endedness that fosters divergent thinking rather than premature closure.
Table 2: Ablation study of AlphaContext on multiple evaluation metrics. Details are the same as Table 1. Methods
Coherence (↑)%
Relevance (↑)%
Engagement (↑)%
Significance (↑)%
Concreteness (↑)%
Uncertainty (↑)%
Diverse Verbs (↑)%
-w/o HOP -w/o MCG -w/o ECO
77.96 74.38 75.62
70.20 71.80 70.57
76.85 72.17 71.80
63.55 65.76 64.53
70.69 69.09 68.72
76.11 71.92 70.69
94.25 93.79 93.36
AlphaContext
81.28
79.06
79.93
71.06
75.49
80.30
96.06
Table 3: Text similarity to expert-designed contexts measured by ROUGE-1, ROUGE-L, and BERTScore. Higher is better.
5.2
Experimental Results and Analysis
BERTScore Main Performance Evaluation (To Q1). We com(↑)%
ROUGE-1 (↑)%
ROUGE-L (↑)%
DeepSeek-V3.1 Qwen3-235B-A22B Llama3.3-70B-Instruct LongWriter-llama3.1-8b LongWriter-glm4-9b GPT-5.1 Gemini-3.0-Pro-Preview DOC-v2 CRITICS SS-GEN
26.22 20.03 20.23 19.88 24.42 15.25 22.89 18.24 17.83 27.80
20.53 16.42 16.32 16.33 20.34 12.46 18.59 15.65 15.14 21.33
80.94 80.39 80.81 81.42 81.33 79.28 80.31 79.87 79.74 81.07
AlphaContext
30.41
25.48
81.88
Methods
in tables since they are consistently small.
These psychometric dimensions were iteratively refined through extensive reviews by senior experts in creativity psychology and aligned with established standards from the Future Problem Solving Program (FPSP) (Torrance et al., 1976), which supports strong content validity. Following arena-hardauto (Li et al., 2025b), we use contexts generated by DeepSeek-V3.1 as the reference baseline and obtain quantified scores through pairwise comparisons with other generated contexts. We also report Diverse Verbs (Fan et al., 2019) to measure action diversity in the context. The prompt template is provided in Appendix I.2. We also report ROUGE-1, ROUGE-L, and BERTScore to measure similarity to expert-designed contexts. Empirically, our real-world human study further supports construct validity by showing significant positive correlations with standardized creativity measures. Implementation Details. All open-source models are locally deployed and run with vLLM on 8× NVIDIA H200 GPUs. For DOC-v2, CRITICS, and SS-GEN, DeepSeek-V3.1 is used as the generation engine. Additionally, DeepSeek-V3.1 serves as the evaluator model for all judgments. To mitigate position bias in pairwise judgments, we evaluate each context pair twice with swapped orders, and repeat this procedure for two rounds, resulting in four evaluations per pair. We omit standard deviations
pare AlphaContext with 10 baselines on CreaTE to assess multi-dimensional context quality for creativity assessment. Diverse Verbs is computed automatically, while Coherence, Relevance, Engagement, Significance, Concreteness, and Uncertainty are judged by an LLM. We follow arena-hardauto (Li et al., 2025b) and conduct pairwise comparisons against the DeepSeek-V3.1 output as the reference model. For each subjective metric, we report the positive rate over the reference, which is set at 50% by definition. Table 1 summarizes the results on CreaTE. AlphaContext ranks first on all seven metrics, with the largest gains on Coherence, Engagement, Significance, and Uncertainty, which are key for constructing coherent stimuli that implicitly cue challenges and elicit open-ended creative thinking. Notably, SS-GEN surpasses GPT-5.1 and Gemini-3.0-Pro-Preview on Significance, while AlphaContext further raises it to 71.06%. This suggests that AlphaContext’s planning, search-based generation, and iterative optimization collectively drive consistent improvements across metrics. Ablation Study (To Q2). To quantify module contributions, we build three ablated variants by removing the HyperTree Outline Planner (HOP), the MCTS-based Context Generator (MCG), and the Evolutionary Context Optimizer (ECO). Table 2 shows that removing any module degrades performance. In particular, removing HOP yields a sharp Relevance drop to 70.20%, indicating that hierarchical planning is critical for keeping assessment cues aligned with the intended theme. Removing MCG lowers Coherence to 74.38% and also reduces Engagement and Uncertainty, suggesting that MCTS search helps preserve long-range structure and the open-endedness needed to elicit creative thinking. Removing ECO decreases all metrics, most notably Uncertainty, showing that MAP-Elites refinement is important for expanding
10
Counts Gauss Fit
1.0
8
Creativity Score Fitted Y 95% Confidence Band
0.8
6
0.6 4
(a) AlphaContext vs. GPT-5.1 (b) AlphaContext vs. GPT-5.1 (Human) (DeepSeek-V3.1)
Figure 3: Preference evaluation of AlphaContext vs. GPT-5.1 under human and DeepSeek-V3.1 judgments.
(a) AlphaContext vs. Gemini (Human)
(b) AlphaContext vs. Gemini (DeepSeek-V3.1)
Figure 4: Preference evaluation of AlphaContext vs. Gemini-3.0-Pro-Preview under human and DeepSeekV3.1 judgments.
stylistic coverage while maintaining assessment cues. This decline arises from two complementary roles of the ECO. First, it expands stylistic coverage by maintaining niche-specific elites in a 3D behavior space, which broadens the diversity of generated contexts. Second, it acts as an effective quality filter through the iterative mutation–evaluation–update loop, which polishes consistency and theme alignment beyond raw MCTS seeds. The ablation results thus validate that the MAP-Elites refinement simultaneously enhances stylistic diversity and core quality while preserving assessment cues. Overall, the ablation results show that the three modules make complementary contributions. Context Similarity Evaluation (To Q3). To assess how closely AlphaContext outputs match expertdesigned contexts given the same inputs, we use 16 expert contexts that were deployed in real creativity assessments and validated by domain experts as references. We compute ROUGE-1, ROUGE-L, and BERTScore, which reflect lexical overlap, longspan matching, and semantic similarity, respectively. Table 3 shows that AlphaContext achieves the best results in all three metrics. It reaches 30.41% ROUGE-1 and 25.48% ROUGE-L, outperforming SS-GEN by 2.61% and 4.15%, and also
0.4
2
0.2
0 0.0 0.2 0.4 0.6 0.8 1.0 Participants' Creativity Scores
0.0 0.0 0.2 0.4 0.6 0.8 1.0 Creativity Score - AlphaContext
(a)
(b)
Figure 5: Human study results for AlphaContext. (a): distribution of participant creativity scores with a Gaussian fit. (b): Pearson correlation between AlphaContextbased scores and AUT (Alternative Uses Task) scores.
surpasses strong LLM baselines in BERTScore. Overall, AlphaContext matches the most closely expert-designed contexts, aligning with its goal of generating psychometrically grounded creativity assessment materials. For more details, please refer to our Appendix B. Preference Evaluation (To Q4). To examine whether the LLM judge aligns with human preferences, we compare AlphaContext with strong baselines (GPT-5.1 and Gemini-3.0-Pro-Preview) via pairwise preference judgments from both human evaluators and DeepSeek-V3.1. To mitigate position bias, each pair is judged twice with swapped orders, and we report win, tie, and lose rates, where win indicates a preference for AlphaContext. Figure 3 and Figure 4 show DeepSeek-V3.1 closely matches human preferences. Against GPT-5.1, AlphaContext wins 62.07% under human judgments and 60.10% under DeepSeek-V3.1, with lose rates around 13%. Against Gemini-3.0-Pro-Preview, AlphaContext wins 73.89% under human judgments and 67.49% with DeepSeek, with low lose rates of 2.96% and 5.91%. The human–LLM agreement is high (Cohen’s κ > 0.8), supporting the reliability of the LLM judgments for our main evaluations. Real-World Human Study (To Q5). We validate the assessment effectiveness of AlphaContextgenerated contexts through a real-world study with 36 secondary-school students. As shown in Figure 5a, the scores exhibit a symmetric unimodal pattern, and the Gaussian fit closely matches the empirical histogram, suggesting a stable measurement behavior. For criterion validity, we compare AlphaContext-based scores with those of the Alternative Uses Task (AUT), a widely used standardized creativity test. The two scores are collected independently, and the Pearson correlation shows
a significant positive association (r = 0.3770), as shown in Figure 5b. Notably, according to standard guidelines on psychology and creativity assessment (Gignac and Szodorai, 2016; Funder and Ozer, 2019; Runco and Acar, 2012; Beaty and Johnson, 2021; Beaty et al., 2022; Benedek et al., 2013), a correlation coefficient of r = 0.3770 is regarded as practically meaningful and provides reasonable support for criterion validity. This result indicates that the creativity levels elicited by our generated contexts are consistent with an established benchmark. Overall, the human study provides real-world evidence that AlphaContext can measure student creativity in authentic educational settings. More details can be found in Appendix D. Case Study (To Q6). To evaluate measurementlevel alignment with expert assessment, we conduct a controlled case study on the same theme using three contexts: expert-designed, AlphaContextgenerated, and Gemini-3-Pro-Preview-generated. We simulate 30 virtual participants with diverse response styles, collect creativity scores, and compare the induced rankings using Spearman correlation and R2 fit. Figure 6 shows that AlphaContext better matches the expert context (rs = 0.84) than Gemini (rs = 0.58), indicating closer outcomelevel consistency with expert-designed assessments. Details are provided in Appendix E.
(a) AlphaContext vs. Expert
(b) Gemini vs. Expert
Figure 6: Case study on measurement-level alignment with expert assessment. Scatter plots compare expertinduced score ranks (x-axis) with ranks induced by generated contexts (y-axis) for 30 simulated participants.
Computational Cost Analysis (To Q7). We analyze the computational cost of AlphaContext in terms of generation time and token consumption, in comparison with baseline LLMs and manual expert design. Table 4 reports the average time and token usage required to generate one creativity context. Although AlphaContext requires more tokens and longer inference time than standard zeroshot prompting, this overhead comes from its full
Table 4: Average generation time and token consumption per context. Method
Time (s)
Tokens (k)
GPT-5.1 Gemini-3.0-Pro-Preview AlphaContext (Ours)
23.50 46.31 226.99
2.36 3.29 12.89
pipeline and is necessary to ensure psychometric validity beyond what simple prompting can reliably provide. In practice, AlphaContext generates one context in about 6 minutes, making it practical for high-quality dataset construction. By contrast, manual expert design typically requires at least one week (Crabbe, 1989; Barbot et al., 2019a). AlphaContext therefore substantially reduces human labor while maintaining quality. In addition, it uses a locally deployed open-source model, avoiding costly closed-source APIs and improving transparency. Overall, the added computational cost is justified by gains in validity, reliability, and automation.
6
Conclusion
This paper introduces AlphaContext, an evolutionary generator for psychometric assessment contexts that integrates rule-guided outline planning, sentence-level MCTS generation, MAP-Elites quality–diversity optimization, and assessment-guided refinement through virtual participant simulation. Across extensive experiments, AlphaContext consistently outperforms strong LLM baselines and structured frameworks on 7 evaluation dimensions, and shows higher alignment with expert-designed contexts. While it consumes more tokens and requires longer generation time than baseline LLMs, its computational overhead is fully acceptable, and it simultaneously achieves significantly higher assessment validity and generation stability. Human–LLM preference evaluations support reliable automated judging, and a real-world study provides practical validity evidence. Overall, AlphaContext offers a scalable way to produce contexts for creativity assessment while reducing the reliance on scarce expert writing. AlphaContext is designed as a context generator for human creativity assessment, while also providing standardized contexts for benchmarking LLM creativity. Current experiments focus on futureoriented contexts and a compact expert-curated input set; extending to broader domains, age groups, and languages is an important direction.
Limitations AlphaContext primarily targets generating psychometrically grounded creativity assessment contexts that are usable in real testing settings. However, achieving stable discourse-level coherence and measurement-relevant cue control relies on sentence-level MCTS and MAP-Elites refinement, which require repeated model calls and scoring. As a result, the overall generation cost depends not only on AlphaContext’s design but also on the underlying LLM and judge configuration, making efficiency comparisons sensitive to the chosen models and evaluation setup. In addition, psychometrically suitable title–theme inputs are still relatively scarce, which can affect the scale of benchmarking and the extent to which conclusions transfer beyond our current future-oriented setting. AlphaContext offers strong controllability and assessment alignment, while direct prompting is typically cheaper but less reliable for measurementoriented constraints. Notably, context generation is not a real-time or online task, and the computational overhead is offset by massive reductions in manual expert effort. In future work, we plan to expand expert-curated inputs and use AlphaContext to produce high-quality training data for finetuning of lightweight generators. This will improve efficiency while preserving assessment utility, enabling broader coverage across domains, populations, and deployment settings.
Ethical Considerations This work introduces the CreaTE dataset for creativity-context generation. The title–theme inputs are authored by creativity-psychology experts under a shared specification and contain no personal information. We screen the dataset to remove sensitive or identifiable content, and we prioritize both data quality and ethical compliance during curation. In particular, we apply strict quality-control procedures, including thorough manual review, to ensure broad coverage while proactively addressing potential bias and sensitivity concerns and we curate the dataset in accordance with applicable privacy and research-ethics standards. Our work also includes a real-world human study to validate the assessment effectiveness of AlphaContext-generated contexts. This study has been reviewed and approved by the Institutional Review Board (IRB) of the affiliated university (IRB Approval No. HR2-0478-2025). Participation was
voluntary, and all participants were informed of the study purpose and procedures, with the right to withdraw at any time without penalty. Before the study, we obtained written informed consent from participants and their guardians, and the consent materials specified the study goals, tasks, potential risks, and data use and protection measures. All collected responses were anonymized by removing personal identifiers, stored securely with restricted access, and reported only in aggregate. The study involved minimal risk, as participants completed open-ended creativity tasks similar to typical classroom activities, and we did not request or record sensitive personal information.
Acknowledgements We would like to thank the anonymous reviewers for constructive comments. This work is supported by the National Natural Science Foundation of China (No. 62476091), the General Program in Education of the National Social Science Fund of China (No. BEA230071), and the Key Program in Education of the National Social Science Fund of China (No. ABA220028).
References T.M. Amabile. 1983. The social psychology of creativity: A componential conceptualization. Journal of Personality and Social Psychology, 45(2):357. T.M. Amabile. 2018. Creativity in context: Update to the social psychology of creativity. Routledge. Minwook Bae and Hyounghun Kim. 2024. Collective critics for creative story generation. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 18784–18819, Miami, FL, USA. Yushi Bai, Jiajie Zhang, Xin Lv, Linzhi Zheng, Siqi Zhu, Lei Hou, Yuxiao Dong, Jie Tang, and Juanzi Li. 2025. Longwriter: Unleashing 10,000+ word generation from long context llms. In Proceedings of the 13th International Conference on Learning Representations, Singapore. B. Barbot, R.W. Hass, and R. Reiter-Palmon. 2019a. Creativity assessment in psychological research:(re) setting the standards. Psychology of Aesthetics, Creativity, and the Arts, 13(2):233. Baptiste Barbot, Richard W Hass, and Roni ReiterPalmon. 2019b. Creativity assessment in psychological research:(re)setting the standards. Psychology of Aesthetics, Creativity, and the Arts, 13(2):233.
R.E. Beaty and D.R. Johnson. 2021. Automating creativity assessment with semdis: An open platform for computing semantic distance. Behavior Research Methods, 53(2):757–780. R.E. Beaty, D.R. Johnson, D.C. Zeitlen, and B. Forthmann. 2022. Semantic distance and the alternate uses task: Recommendations for reliable automated assessment of originality. Creativity Research Journal, 34(3):245–260. Ronald A Beghetto. 2021. There is no creativity without uncertainty: Dubito ergo creo. Journal of Creativity, 31:100005. Ronald A Beghetto and Garrett J Jaeger. 2022. Uncertainty: A catalyst for creativity, learning and development. Creativity Theory and Action in Education, 6. M. Benedek, C. Mühlmann, E. Jauk, and A.C. Neubauer. 2013. Assessment of divergent thinking by means of the subjective top-scoring method: Effects of the number of top-ideas and time-on-task on reliability and validity. Psychology of Aesthetics, Creativity, and the Arts, 7(4):341.
Jérôme Guegan, Julien Nelson, and Todd Lubart. 2017. The relationship between contextual cues in virtual environments and creative processes. Cyberpsychology, Behavior, and Social Networking, 20(3):202– 206. Jiajun Guo and Sarah Woulfin. 2016. Twenty-first century creativity: An investigation of how the partnership for 21st century instructional framework reflects the principles of creativity. Roeper Review, 38(3):153–161. Antonio Laverghetta Jr., Simone Luchini, Averie Linnell, Roni Reiter-Palmon, and Roger E. Beaty. 2024. The creative psychometric item generator: a framework for item generation and validation using large language models. In Proceedings of the 3rd Workshop on Artificial Intelligence and Creativity colocated with 27th European Conference on Artificial Intelligence, pages 59–73, Santiago de Compostela, Spain. Brandon Koh and Angela K-y Leung. 2019. A time for creativity: How future-oriented schemas facilitate creativity. Journal of Experimental Social Psychology, 84:103816.
A.B. Crabbe. 1989. The future problem solving program. Educational Leadership, 7(1):27–29.
Hye-Kyung Lee. 2022. Rethinking creativity: Creative industries, ai and everyday creativity. Media, Culture & Society, 44(3):601–612.
Anne Borland Crabbe. 1982. Creating a brighter future: An update on the future problem solving program. Journal for the Education of the Gifted, 5(1):2–11.
Kuang-Huei Lee, Ian Fischer, Yueh-Hua Wu, Dave Marwood, Shumeet Baluja, Dale Schuurmans, and Xinyun Chen. 2025. Evolving deeper LLM thinking. arXiv preprint, arXiv:2501.09891.
Angela Fan, Mike Lewis, and Yann N. Dauphin. 2019. Strategies for structuring story generation. In Proceedings of the 57th Conference of the Association for Computational Linguistics, pages 2650–2660, Florence, Italy. Xinyu Fang, Zhijian Chen, Kai Lan, Lixin Ma, Shengyuan Ding, Yingji Liang, Xiangyu Zhao, Farong Wen, Zicheng Zhang, Guofeng Zhang, Haodong Duan, Kai Chen, and Dahua Lin. 2025. Creation-mmbench: Assessing context-aware creative intelligence in MLLM. arXiv preprint, abs/2503.14478. Yi Feng, Mingyang Song, Jiaqi Wang, Zhuang Chen, Guanqun Bi, Minlie Huang, Liping Jing, and Jian Yu. 2025. SS-GEN: A social story generation framework with large language models. In AAAI-25, Sponsored by the Association for the Advancement of Artificial Intelligence, pages 1300–1308, Philadelphia, PA, USA. D.C. Funder and D.J. Ozer. 2019. Evaluating effect size in psychological research: Sense and nonsense. Advances in Methods and Practices in Psychological Science, 2(2):156–168. G.E. Gignac and E.T. Szodorai. 2016. Effect size guidelines for individual differences researchers. Personality and Individual Differences, 102:74–78.
Jiaming Li, Yukun Chen, Ziqiang Liu, Minghuan Tan, Lei Zhang, Yunshui Li, Run Luo, Longze Chen, Jing Luo, Ahmadreza Argha, Hamid Alinejad-Rokny, Wei Zhou, and Min Yang. 2025a. STORYTELLER: an enhanced plot-planning framework for coherent and cohesive story generation. In Findings of the Association for Computational Linguistics, pages 20818– 20846, Vienna, Austria. Tianle Li, Wei-Lin Chiang, Evan Frick, Lisa Dunlap, Tianhao Wu, Banghua Zhu, Joseph E. Gonzalez, and Ion Stoica. 2025b. From crowdsourced data to highquality benchmarks: Arena-hard and benchbuilder pipeline. In Forty-second International Conference on Machine Learning, Vancouver, BC, Canada. Luchini, Simone A, Maliakkal, Nadine T, DiStefano, Paul V, Laverghetta Jr, Antonio, Patterson, John D, Beaty, Roger E, and Roni Reiter-Palmon. 2025. Automated scoring of creative problem solving with large language models: A comparison of originality and quality ratings. Psychology of Aesthetics, Creativity, and the Arts. Aidan Mclaughlin, James Campbell, Anuja Uppuluri, and Yiming Yang. 2024. Aidanbench: Stress-testing language model creativity on open-ended questions. In NeurIPS 2024 Workshop on Language Gamification.
Peter Meusburger. 2009. Milieus of creativity: The role of places, environments, and spatial contexts. Milieus of creativity: An interdisciplinary approach to spatiality of creativity, 2:97–153. Michael D Mumford, Robert Martin, Samantha Elliott, and Tristan McIntosh. 2018. Creative thinking in the real world. The nature of human creativity, pages 147–65. Alexander Novikov, Ngân Vu, Marvin Eisenberger, Emilien Dupont, Po-Sen Huang, Adam Zsolt Wagner, Sergey Shirobokov, Borislav Kozlovskii, Francisco J. R. Ruiz, Abbas Mehrabian, M. Pawan Kumar, Abigail See, Swarat Chaudhuri, George Holland, Alex Davies, Sebastian Nowozin, Pushmeet Kohli, and Matej Balog. 2025. Alphaevolve: A coding agent for scientific and algorithmic discovery. arXiv preprint, arXiv:2506.13131. Shawn M Okuda, Mark A Runco, and Dale E Berger. 1991. Creativity and the finding and solving of realworld problems. Journal of Psychoeducational assessment, 9(1):45–53. Kai Ruan, Xuan Wang, Jixiang Hong, and Hao Sun. 2024. Liveideabench: Evaluating llms’ scientific creativity and idea generation with minimal context. arXiv e-prints, pages arXiv–2412. M.A. Runco and S. Acar. 2012. Divergent thinking as an indicator of creative potential. Creativity Research Journal, 24(1):66–75.
Xueyang Wang, Wei Liu, Kaixiang Zhuang, Cheng Liu, Jingyi Zhang, Li Fan, Qunlin Chen, and Jiang Qiu. 2025a. Neural representations of noncentral events during narrative encoding predict subsequent story ending originality. Science Advances, 11(17):eadu5251. Yixuan Wang, Jiale Feng, Yue Huang, Xuruo Pan, Zhongjing Huang, Zhi Liu, and Hong Qian. 2025b. A style-aware polytomous diagnostic model for individual traits. In Proceedings of the 28th European Conference on Artificial Intelligence, pages 2698– 2705, Bologna, Italy. Hongqiu Wu, Weiqi Wu, Tianyang Xu, Jiameng Zhang, and Hai Zhao. 2025a. Towards enhanced immersion and agency for llm-based interactive drama. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics, pages 11166– 11182, Vienna, Austria. Shirley Wu, Michel Galley, Baolin Peng, Hao Cheng, Gavin Li, Yao Dou, Weixin Cai, James Zou, Jure Leskovec, and Jianfeng Gao. 2025b. Collabllm: From passive responders to active collaborators. In Proceedings of the 42nd International Conference on Machine Learning, Vancouver, Canada. Kevin Yang, Dan Klein, Nanyun Peng, and Yuandong Tian. 2023. DOC: improving long story coherence with detailed outline control. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics, pages 3378–3465, Toronto, Canada.
Mark A Runco and Garrett J Jaeger. 2012. The standard definition of creativity. Creativity research journal, 24(1):92–96.
Appendix
Mark A Runco, Burak Turkman, Selcuk Acar, and Mustafa V Nural. 2017. Idea density and the creativity of written works. Journal of Genius and Eminence, 2(1):26–31.
AlphaContext takes a title and a theme as input, so evaluation requires inputs that are explicitly designed for creativity assessment. Existing storygeneration datasets are ill-suited for this purpose: they target narrative completeness and stylistic richness, but do not ensure that the topic focus, implicit cues, and task intent satisfy psychometric requirements. Using such datasets would confound the evaluation, as poor performance could arise from mismatched inputs rather than the generation method. To address this gap, we curate CreaTE, a compact yet high-quality input set for creativity-context generation. CreaTE contains 203 title–theme pairs spanning diverse domains, balancing evaluation cost with broad coverage. Each entry is authored by three creativity-psychology experts under a shared specification: the title provides a concrete anchor, while the theme delineates the intended problem space and key creative tensions to be elicited, as shown in Figure 7. A subset of expert-designed
Chenglei Si, Diyi Yang, and Tatsunori Hashimoto. 2025. Can llms generate novel research ideas? A largescale human study with 100+ NLP researchers. In Proceedings of the 13th International Conference on Learning Representations, Singapore. Robert J Sternberg. 1984. Toward a triarchic theory of human intelligence. Behavioral and Brain Sciences, 7(2):269–287. Robert J Sternberg and Todd I Lubart. 1999. The concept of creativity: Prospects and paradigms. Handbook of creativity, 1(3-15). E. P. Torrance, C. B. Bruch, and J. P. Torrance. 1976. Interscholastic futuristic creative problem-solving. The Journal of Creative Behavior, 10(2):117–125. Tyler J VanderWeele. 2025. Intellectual and viewpoint diversity: Importance, scope and bounds. Education Sciences, 15(12):1592.
A
CreaTE Dataset
titles and themes is adapted from publicly available FPSP topic resources1 . We refine the dataset through cross-checking and iterative revision, and include an entry only after expert consensus on assessment relevance, clarity, and correctness, with edits applied to remove ambiguity or unintended cues. Finally, we conduct compliance screening to ensure that no sensitive or identifiable information is present. [ { "title": "Youth in Competitive Sports", "theme": "Youth competitive sports: performance pressure, mental health, injury risk, equity of access, and technology-enhanced training." }, { "title": "Nanotechnology in Daily Life", "theme": "Nanotechnology: smart materials, medicine, environmental cleanup, manufacturing change, and long-term safety and regulation." }, …… { "title": "Neurotechnology Future Scene", "theme": "Medical Rehabilitation-Focused Neurotechnology: Rehab applications, state ethics review, consent, brain data security." }, ]
Figure 7: Example input format of CreaTE. Each entry includes a title and a theme, designed to support psychometrically grounded creativity context generation.
B
Preference Evaluation
We provide additional details on the human evaluation protocol and inter-annotator agreement analysis for the preference study involving Gemini-3.0Pro-Preview. For human judgments, we recruited three evaluators and adopted a standardized rating protocol with a pre-study calibration session. The human evaluation checklist is strictly aligned with our metric definitions (Coherence, Relevance, Engagement, Significance, Concreteness, Uncertainty) to ensure consistent interpretation. Inter-rater agreement among the three evaluators meets the required consistency standard, and we report the mean of their judgments as the final human preference result. Consistent with observations in the main text, DeepSeek’s judgments closely track human preferences, with high human–LLM agreement (Cohen’s κ > 0.8), further validating the reliability of automated evaluation in our experiments. 1
https://fpspi.org/topics/
C
Cue Coverage and Diversity in a Case Comparison
Under the same input theme, AlphaContext produces a context fragment with noticeably broader assessment-cue coverage than SS-GEN. Figure 8 shows that AlphaContext can surface multiple challenge dimensions within a single coherent narrative move, so the scenario invites reasoning from several angles rather than focusing on only one. In contrast, SS-GEN typically centers the context around a single dominant cue, which yields a narrower cue footprint and fewer directions for subsequent idea exploration. This qualitative comparison aligns with our design goal: by planning cue placement at the outline level and enforcing outline-grounded generation, AlphaContext increases the diversity of assessment-relevant cues while keeping them implicitly integrated into the story, better supporting creativity assessment that aims to elicit open-ended and multi-perspective thinking. AlphaContext The community organizer, Sarah, chimed in, 'We can also mobilize more volunteers and raise funds through local campaigns and partnerships with businesses. ["Social Relationships", "Economics", "Business & Commerce"].
SS-GEN In the turquoise waters off the coast of Palau, a delicate ecosystem faces an invisible threat. Coral reefs that have thrived for millennia now show signs of bleaching and stress. ["Environment"].
Figure 8: Case comparison under the same input theme: AlphaContext embeds a broader and more diverse set of assessment cues in a coherent fragment, whereas SSGEN concentrates on a single dominant cue, resulting in narrower cue coverage.
D
Detailed Analysis of Human Study
To complement the aggregate Pearson correlation reported in the Real-World Human Study, we further examine how specific creativity dimensions align between AlphaContext and a standardized benchmark using Spearman rank correlation. We used Pearson correlation in the main text because the aggregated total scores empirically exhibit an approximately unimodal, near-Gaussian distribution, making Pearson an appropriate summary of linear association. In contrast, dimension-level creativity scores can be more heterogeneous in both
cognitive mechanisms and distributional shapes. We therefore adopt Spearman correlation to provide a rank-based and more robust analysis of whether the two assessments preserve consistent relative ordering at a finer granularity. Before the formal study, all participants completed the assessment under a standardized administration procedure to ensure comparability across individuals. The instructions, task materials, and time constraints were fixed and delivered in a consistent format. Participants were guided to complete the tasks independently and were discouraged from discussion or external assistance during the session. To protect privacy, we removed personally identifiable information from all collected records, used anonymized participant identifiers for subsequent analysis, and stored the data in an accessrestricted manner. Only de-identified responses and scores were used for reporting and correlation analysis. All human scores used in this analysis were produced under a standardized expert rating protocol. Specifically, three psychology experts served as raters. A domain specialist first developed a detailed scoring rubric and dimension definitions, and then conducted fine-grained training for the raters before annotation. The training session explained each dimension with concrete guidance and calibration examples, ensuring a shared understanding of the scoring criteria and common failure cases. During scoring, the three raters evaluated responses independently. We then computed inter-rater agreement to verify consistency, and the agreement met the required standard. Finally, we report the mean score across the three raters as the human score for each dimension.