LLM-Assisted Augmented Intelligence for Context-Aware Decision

Support: Current Trends and Integrated Approach

Alexander Smirnov

, Andrew Ponomarev

, Nikolay Shilov

and Tatiana Levashova

St. Petersburg Federal Research Center of the Russian Academy of Sciences (SPC RAS),

St. Petersburg, Russian Federation

Keywords: Decision Support, Augmented Intelligence, Large Language Models, Generative AI, Conversational AI,

Dialogue-Based DSS, Evaluative AI.

Abstract: The growing complexity of technical, social, and business systems created and managed by humans determine

the need for effective decision support. Recent advancements in AI push the boundary of what can be

accomplished using AI tools and what are possible modes of human-AI interaction, bringing a concept of

augmented intelligence, extending intellectual capabilities of human by variety of AI-based tools, while

leaving final decision-making (as well as some other operations, e.g., goal-setting, coordination, control) to a

human. This paper explores possibilities of using augmented intelligence for decision support. Starting with

a general structure of decision-making process, it highlights and reviews current trends in several branches of

AI, that are most important for decision support. Then, it proposes an integrated approach combining

conversational, generative, and evaluative AI. Distinguishing features of the proposed approach are

integration and mutual enrichment of data- and model-based techniques, as well as using modern LLMs as a

basis for human-AI interaction during decision-making.

1 INTRODUCTION

The growing complexity and number of tasks solved

by humans determine the need for the design of

effective decision support systems (DSSs). Over the

last few decades, a variety of techniques and

approaches powering DSSs have been proposed and

explored (Megawaty and Ulfa 2020). Recently, the

development of machine learning methods, and in

particular, large language models (LLMs), has led to

their introduction into various human activity

processes, including decision-making processes.

However, this trend is also associated with a number

of problems, such as the complexity of interaction

with such models, the low level of trust in them, the

unpredictability of the results they generate. This

highlights the relevance and global nature of the tasks

considered in the paper.

Among the approaches aimed at addressing these

issues, one can highlight the development of

https://orcid.org/0000-0001-8364-073X

https://orcid.org/0000-0002-9380-5064

https://orcid.org/0000-0002-9264-9127

https://orcid.org/0000-0002-1962-7044

dialogue-based DSSs, which involve both the

argumentation of alternative solutions presented to

the user and their evaluation alongside the assessment

of user-proposed solutions, ultimately aiding the user

in reaching a final decision through step-by-step

recommendations. Therefore, there is urge in the

development of new methods that would allow both

to synthesize new generation of DSSs and to use

them, supplementing human intelligence with

artificial intelligence to improve the effectiveness of

decision-making processes by preserving the leading

role of humans in the decision-making process and

organizing constructive dialogue between the

decision maker (DM) and AI.

Context awareness is critical for effectiveness of

decision support, as decisions inherently depend on

situational factors, user constraints, and domain-

specific knowledge. Without proper context

integration, even technically optimal solutions may

Smirnov, A., Ponomarev, A., Shilov, N. and Levashova, T.

LLM-Assisted Augmented Intelligence for Context-Aware Decision Support: Current Trends and Integrated Approach.

DOI: 10.5220/0013714300004000

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 17th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2025) - Volume 2: KEOD and KMIS, pages

409-416

ISBN: 978-989-758-769-6; ISSN: 2184-3228

409

Figure 1: Decision-making cycle and relevant areas of AI.

prove impractical. Thus, context awareness serves not

as an enhancement but as a core requirement for DSS

in complex, real-world environments. Context-

dependent decision support focuses on formulating

solutions based on context (the task status, the

situation of the DM, as well as her/his preferences),

and a DSS implementing this approach should

include the following stages: (1) clarifying the current

task's formulation, (2) generating solutions feasible

within the current context, (3) evaluating these

solutions, including explanations, and providing the

DM with recommendations for making the final

decision.

Integrating the DM's intelligence with AI models

and methods can significantly improve the

effectiveness of context-dependent decision support.

Such integration, including the development of AI

models designed to support and enhance human

capabilities (as opposed to autonomous AI decision-

making), is referred to as augmented intelligence.

Implementation of the aforementioned stages of

context-dependent decision support requires:

(1) dialogue with the DM to clarify the task (since the

user's preferences and task conditions are rarely fully

known initially), (2) generation of possible solutions

(alternatives), and (3) their evaluation while ensuring

trust between the DM and the decision-supporting AI.

This trust is achieved, in particular, through

explainable reasoning. To meet these requirements, it

is advisable to use methods from conversational,

generative, and evaluative AI, respectively (Fig. 1).

This can be viewed as a natural elaboration of existing

trends in human-AI collaborative decision support

and (Smirnov, Shilov, and Ponomarev 2022).

Thus, the approach proposed in the paper focuses

on the three aforementioned types of AI

(conversational, generative, and evaluative) in terms

of developing models and methods necessary for their

effective use by DMs within the framework of

augmented intelligence. Application of LLMs

appears promising for implementing these.

The rest of the paper is organized as follows.

Section 2 provides a brief review of existing research

in areas, relevant to the proposed approach. Based on

this review, Section 3 summarizes key principles of

the approach, which is presented in more detail in

Section 4.

2 RELATED WORK

This section is structured according to the four main

research directions indicated above, namely: LLM-

based decision support, conversational, generative,

and evaluative AI.

2.1 LLM-Based Decision Support

In recent years, with the rapid advancement of LLMs,

a variety of general (and sufficiently universal)

approaches have been proposed to support decision-

making through natural language interaction with

users.

There are several approaches trying to integrate

different types of AI. For example, the integrated

methodology proposed in (Ramaul, Ritala, and

Ruokonen 2024) emphasizes the combined use of

generative and conversational AI in chatbots like

ChatGPT. The methodology is based on cyclic

bidirectional interactions between these two AI types:

the human initiates the generative AI, which then

engages the conversational AI for further dialogue.

This allows users to maintain continuous

conversation with the system, initiating new queries,

receiving responses, requesting clarifications or

refinements, and obtaining updated answers. Similar

Conversational

Generative

Evaluative

Problem

clarification

Identification

of alternatives

Evaluation of

the alternatives

LLM

problem

model

KMIS 2025 - 17th International Conference on Knowledge Management and Information Systems

410

methodologies, integrating several types of AI are

Ask Grapho (Gutierrez de Esteban 2023), DeLLMa

(Liu et al. 2024) and mPulse methodology (Danzer

2024).

From the technical perspective, using LLM for

decision support is typically done by enriching LLM

with additional modules and tools (sometimes, this is

referred to as LLM-based agents). For example,

RAGADA architecture (Pitkäranta and Pitkäranta

2024) integrates a library of specialized methods and

algorithms with the generative capabilities of LLMs,

enhanced by Retrieval-Augmented Generation

(RAG) technology, providing means to select

appropriate problem-specific algorithms and present

their outputs in natural language.

Beyond general-purpose DSS architectures

employing LLMs for user interaction (among other

functions), there are also specialized frameworks for

particular DSS categories, e.g., recommender

systems, and conversational recommender systems

(Feng et al. 2023).

A distinctive feature of the approach proposed in

this paper is bidirectional integration of LLMs and

conceptual modeling – leveraging conceptual models

to enhance decision support quality while

simultaneously employing LLMs for conceptual

model construction during DSS development. Initial

attempts to apply LLMs for building domain

conceptual models have already emerged. For

instance, (Kommineni, König-Ries, and Samuel

2024) presents an automated approach to knowledge

graph construction using LLMs. Similar efforts

appear in (Caufield et al. 2023; Babaei Giglou,

D’Souza, and Auer 2023; Lam et al. 2024).

While LLMs may compete with ontologies in

certain applications, their combined use shows

significant potential. Moreover, complete automation

of high-quality ontology development through LLMs

remains unachievable (Neuhaus 2023), necessitating

human-AI collaboration (through dialogue and DM

involvement).

2.2 Generative AI

Generative Adversarial Networks (GANs) (Kusiak

2025; Chakraborty et al. 2024) represent one of the

most powerful deep learning methods for artifact

generation. Contemporary generative models are

capable of producing diverse artifacts including text,

images, videos, tabular data, and parameterized

graphs (Fan and Huang 2019; Fan, Tech, and Huang

2019; Zhou et al. 2019; Jia et al. 2023). These models

can be enhanced through conditional architectures

that incorporate additional input parameters

(conditional GAN), e.g, (Fan, Tech, and Huang

2019), and specialized modules that optimize training

efficiency for domain-specific constraints (De Cao

and Kipf 2018). Such capabilities make them

potentially suitable for generating feasible solutions

in DSSs, where outputs must satisfy both explicit

requirements and implicit patterns learned from

training data.

Notably, generative AI extends beyond neural

network implementations. Alternative approaches

include knowledge-based systems employing

symbolic knowledge representations (dynamic

modeling, metaheuristics, constraint satisfaction) and

evolutionary algorithms (genetic, swarm intelligence,

cellular automata) (Liao et al. 2024; Jiang et al. 2024).

Hybrid methodologies also show promise, such as the

integrated approach combining metaheuristics with

question generation to create educationally balanced

assignments with comprehensive topic coverage

(Láng and Dömsödi 2024).

2.3 Conversational AI for Decision

Support

Within the domain of conversational AI for context-

aware DSS, two key challenges merit particular

attention: model-oriented dialogue management

using LLMs for situation modeling and requirement

refinement, and LLM adaptation for domain-specific

dialogues.

The refinement of user requirements through

dialogue represents a critical application of LLMs in

DSS. Here, the primary function of LLMs involves

interpreting user inputs and mapping them to

elements of the task model, including their potential

values (Lawless et al. 2024; Han et al. 2023).

However, in most implementations, the task model is

predefined, with LLM integration designed

accordingly. Extending these approaches to work

with generalized structural models could significantly

streamline DSS development.

Domain adaptation of LLMs aims to enhance

response accuracy and reduce hallucinations. Widely

applied adaptation methods fall into two broad

categories: parametric and non-parametric.

Parametric adaptation involves fine-tuning model

parameters using domain-specific training samples.

This process not only improves the model's grasp of

specialized terminology and conceptual relationships

but also optimizes dialogue performance for specific

tasks (Tu et al. 2025). The complexity of parametric

adaptation, coupled with the inherent optimization of

many LLMs for general dialogue and instruction-

following, has led to widespread adoption of non-

LLM-Assisted Augmented Intelligence for Context-Aware Decision Support: Current Trends and Integrated Approach

411

parametric methods. These techniques, collectively

termed “prompt engineering”, augment queries with

relevant information unavailable during initial

training (e.g., post-training updates or proprietary

knowledge). Dozens of prompt engineering

techniques now exist, serving not only for domain

adaptation but also for generating step-by-step

reasoning, reducing hallucinations and other tasks

(Sahoo et al. 2024). Such methods have proven

particularly effective in medical DSS prototypes,

where they incorporate clinical guidelines

(unavailable during model training) into physician

recommendations (Zhao, Wang, and Peng 2024).

2.4 Evaluative AI

The role of evaluative AI within the proposed

approach lies in assessing both human-proposed and

AI-generated solutions, subsequently presenting

these evaluations to DMs alongside explanatory

rationale to facilitate more informed final decisions.

The evaluative AI embodies a paradigm shift

from recommendation-driven to hypothesis-driven

(Le, Miller, Zhang, et al. 2024; Miller 2023), where

the AI system enhances rather than dictates decision-

making processes. Specifically, it aims to strengthen

decision-maker autonomy (Le, Miller, Sonenberg, et

al. 2024), contextual awareness (Le 2023; Le, Miller,

Sonenberg, et al. 2024), and DM’s procedural control

through her/his hypothesis incorporation. While

implementation approaches vary significantly, a

defining characteristic of such evaluative AI systems

is their capacity to present balanced arguments

supporting and opposing each alternative.

Application of evaluative AI in DSS is a

promising research direction, necessitating further

work in three key areas: task model specification,

solution evaluation, and their presentation to DMs.

2.5 Summary

1. The development of augmented intelligence-

based DSS necessitates a comprehensive

methodology that combines: natural language

contextual interaction with users, generation of

appropriate responses or solutions aligned with

user requests, and justification & evaluation of

proposed solutions preserving the final decision

authority with human DMs. The most promising

AI components for such methodology appear to

be generative AI, conversational AI, and

evaluative AI. Currently, no existing

methodology integrates all three AI types for

augmented decision support. While some

approaches combine generative and

conversational AI - where generative AI utilizes

LLMs to formulate responses derived from

logical inference or computational problem-

solving, and conversational AI manages user

interaction through LLMs and NLP techniques.

The proposed in this paper approach aims to

incorporate all three AI types to implement

Simon's decision-making model (Simon 1979).

2. General-purpose LLMs, which form the

foundation of conversational AI implementations,

exhibit several inherent limitations. These

limitations can be effectively addressed by

employing LLMs as components within

integrated solutions supplemented by decision-

making methodologies and specialized reasoning

tools.

3. Current research notably underrepresents the

integration of conceptual modeling with LLMs,

making this an exceptionally promising direction

for advancing conversational AI techniques.

4. The predominant specialization of existing

artifact generation methods (generative AI)

similarly suggests the value of developing more

universal solutions grounded in domain

conceptual models, which would enable broader

applicability with minimal adaptation.

5. The incorporation of evaluative AI into DSS

represents another promising research avenue

demanding development of task model

specification methods, solution evaluation

frameworks, and presentation models for DMs.

3 FOUNDATIONAL PRINCIPLES

This section summarizes key principles that shape the

integrated approach, proposed in the paper. These

principles are based on the literature review, in

particular, on the identification of open issues and

their potential solutions. The principles are structured

along three dimensions: 1) the role of the prospective

DSSs in management activities, their scope and

functionality, 2) mode of interaction with the user

(decision-maker), 3) problem representation and

processing. These dimensions capture both “external”

image of the DSS in its location among other tools

available to the decision-maker, and its “internal”

organization.

KMIS 2025 - 17th International Conference on Knowledge Management and Information Systems

412

Scope and Role

1. DSS is aimed at helping decision-maker to

understand situation in detail, it may propose

solutions, help to evaluate solutions (proposed

either by decision-maker, or by itself), but the

final decision is made by human and human is

responsible for it. Hence, it is an exemplar of

augmented intelligence, when decision-makers

competency and understanding of the problem

situation is extended by external information and

reasoning capabilities of AI.

2. DSS memorizes the scenarios, results of their

executions, and decisions made and can use this

knowledge to make recommendations in the

future.

Interaction with the User

1. DSS interprets problems presented in natural

language. This interface can be supplemented

with other suitable forms of information

presentation and input (depending on the problem

domain of the DSS), but interface in natural

language is obligatory. Despite the ambiguity

immanent to natural language, the support of it as

a main communication media obviates the need

for creating problem-specific user interfaces and

simplifies the adoption of the DSS.

2. DSS interacts with human, when necessary, e.g.,

to refine the query, to agree on the problem model,

etc. In particular, it can help to resolve ambiguities

unavoidable when using natural language

communication. But moreover, problem is rarely

clearly specified in advance, often even decision-

maker a priori doesn’t know all the details and

preferences.

3. AI provides explanations of recommendations

(proposed decisions) and evaluations (of the

decisions proposed by itself or by decision-

maker), giving a traceable critique, allowing to

check whether these results can be trusted.

4. AI infers, manages, and takes into account

decision-makers’ preferences (explicit and

implicit).

Problem Representation and Processing

1. DSS is context-aware, it infers and takes into

account environment of the problem, decision-

makers perspective and other transient

information related to the problem.

2. DSS builds a model of the problem (or several

semantically interoperable models). Specifics of

these models may vary depending on the problem

domain and problem itself, however, model

building on the one hand, allows clarifying

intricacies of the problem at hand and its relations

to other elements of the domain, on the other

hand, it sometimes allows employing efficient and

explainable techniques, fulfilling the

explainability principle. This is one of the most

important principles and in severely influences

several other.

3. DSS decomposes complex problems into simpler

ones. Such decomposition is often made based on

the model of the problem and on some process-

based representation of similar problems.

4. Multi-aspect semantically consistent

representation of the proposed solution. A

potential solution sometimes has to address the

requirements of several stakeholders, who may

deal with the problem on different levels of

abstraction or from different perspectives.

Solution representation should allow such

multiple views, ensuring that they are consistent.

5. Utilization of existing knowledge. Such

utilization can take two forms: on a meta-level,

domain models can be viewed as a refined

representation of domain knowledge. On the

factual level, domain knowledge is represented in

variety of structured and unstructured resources

that can be interpreted and leveraged.

4 THE PROPOSED APPROACH

The article primarily examines support in solving

complex tasks that cannot be fully delegated to AI.

Accordingly, the main operational scenarios of the

decision support system (DSS) are those in which AI

generates recommendations, provides the necessary

information for decision-making (including

explanations), and evaluates possible solutions, while

humans use the AI's outputs to form their own

decisions and select the final solution.

The central element of the proposed approach (see

Fig. 2) for developing context-dependent decision

support systems is the process of model specification

using conceptual modeling techniques, as well as

information extraction and generalization methods to

systematize knowledge about the problem domain.

Thus, according to the proposed methodology,

during the DSS development phase, two closely

interconnected parallel processes are carried out in a

dialog mode: (a) the analysis of unstructured (textual)

and structured (database) information sources about

the problem domain; (b) the construction and

LLM-Assisted Augmented Intelligence for Context-Aware Decision Support: Current Trends and Integrated Approach

413

refinement of a conceptual model of the problem

domain, which includes the main object classes, their

relationships, the decision-maker's (DM) task models,

their possible decomposition into subtasks, and the

anticipated solution scenarios.

During the DSS operation phase, the DM's goal is

identified, and the model of the current task is

specified (populated with concrete attributes). This

refined task model is then used to generate and

evaluate solutions using developed generative and

evaluative AI methods, respectively.

As it was already noted, the proposed concept

amalgamates several branches (or, flavors) of AI,

here, the roles of different types of AI in the proposed

framework are summarized:

1. Generative AI. Within the proposed approach,

the objective of generative AI methods is to construct

solutions aligned with the decision-maker's (DM)

task. A common feature of the generative AI methods

considered here is that they take as input a task model

constructed through user interaction using an LLM

(Large Language Model), along with elements of the

problem domain model. Two key directions in

generative AI are deemed relevant: a) optimization-

based approaches (potentially constraint-aware),

primarily relying on metaheuristics; b) generative

neural network models. To select appropriate

solution-generation methods, it is necessary to

develop a method for determining dependencies

between problem domain models and the

corresponding solution-generation techniques. When

considering solution generation directly via LLMs,

attention should be given to investigating their

analogy with crowd computing. In both cases,

solutions are obtained through unreliable agents, and

crowd computing has developed various techniques to

process such results and enhance their reliability.

2. Conversational AI. The role of conversational

AI methods in the proposed approach is to construct

and refine the problem domain model (during DSS

development) and the specific task model (during DSS

operation) through natural language dialogue with the

user. The proposed approach centers conversational

AI around LLMs, which currently represent the most

promising tool for natural language interaction.

Specifically, LLMs are used for: a) acquiring domain

knowledge, context, and user preferences through

dialogue; b) providing the user with information (e.g.,

solution evaluations) in natural language. However,

generic LLMs are not always suitable for decision

support in specific domains. Therefore, one of two

kinds of adaptation of LLM to the domain can be used:

- parametric adaptation, which involves fine-

tuning the LLM to optimize dialogue behavior for the

given task. The most demanded here are resource-

efficient techniques allowing to reduce computational

complexity of the LLM fine-tuning (e.g., Quantized

Low-Rank Adaptation – QLoRA).

Figure 2: General scheme of the approach.

Data sources

(DBs, text, etc.)

DSS

develope

Domain model

building

DSS development

Abstract

domain

model

Task model

building

Selection and

application of

the specialized

decision support

method

Knowledge on

methods of task

solving

Decision-make

DSS operation

Usin

decisions to im

rove models, reasonin

and further decisions

Personalization and

context

RAG variants

KMIS 2025 - 17th International Conference on Knowledge Management and Information Systems

414

- non-parametric adaptation, which encompasses

solutions implemented via LLM extensions (often

termed LLM agents), including: prompt

engineering methods for structured, goal-oriented

dialogue; retrieval-Augmented Generation (RAG) to

enhance responses with external knowledge, reducing

hallucinations and improving answer quality and

some other techniques.

3. Evaluative AI. The role of evaluative AI

methods in the proposed approach is to assess

solutions—whether proposed by the DM or generated

by AI—and present these evaluations to the DM

along with explanations to support more informed

final decisions. Evaluations are performed using the

task model and problem domain model,

supplemented with pros and cons for each assessed

solution.

5 CONCLUSIONS

The growing complexity of modern business and

technical systems demands advanced decision-

support methodologies that augment human

intelligence rather than replace it. This paper has

explored how conversational, generative, and

evaluative AI—integrated into a unified

framework—can enhance decision-making while

preserving human agency in critical functions like

goal setting and validation. By combining data-driven

and model-based techniques with LLM-mediated

interaction, the proposed approach enables adaptive,

context-aware support that evolves alongside

dynamic real-world challenges.

Key advantages include the system’s ability to:

(1) leverage structured and unstructured knowledge

through hybrid AI methods, (2) provide explainable,

auditable reasoning via evaluative components, and

(3) maintain natural human oversight through

conversational interfaces. Future work should address

computational efficiency and domain-specific

customization while maintaining ethical alignment.

As AI capabilities progress, such augmented

intelligence systems will become indispensable for

balancing automation with human judgment in high-

stakes decision environments.

ACKNOWLEDGEMENTS

The research is funded by the Russian Science

Foundation (grant number 25-11-00127,

https://rscf.ru/project/25-11-00127/).

REFERENCES

Alvarez-Melis, David, Harmanpreet Kaur, Hal Daumé,

Hanna Wallach, and Jennifer Wortman Vaughan. 2021.

“From Human Explanation to Model Interpretability: A

Framework Based on Weight of Evidence.” In 9th AAAI

Conf. on Human Computation and Crowdsourcing

(HCOMP 2021). https://arxiv.org/abs/2104.13299.

Babaei Giglou, Hamed, Jennifer D’Souza, and Sören Auer.

2023. “LLMs4OL: Large Language Models for

Ontology Learning.” In The Semantic Web – ISWC

2023, 408–27. https://doi.org/10.1007/978-3-031-

47240-4_22.

Cao, Nicola De, and Thomas Kipf. 2018. “MolGAN: An

Implicit Generative Model for Small Molecular

Graphs.” In ICML 2018 Workshop on Theoretical

Foundations and Applications of Deep Generative

Models. https://doi.org/10.48550/arXiv.1805.11973.

Castelnovo, Alessandro, Riccardo Crupi, Nicolò Mombelli,

Gabriele Nanino, and Daniele Regoli. 2024.

“Evaluative Item-Contrastive Explanations in

Rankings.” Cognitive Computation, July. https://doi.

org/10.1007/s12559-024-10311-2.

Caufield, J. Harry, Harshad Hegde, Vincent Emonet, Nomi

L. Harris, Marcin P. Joachimiak, Nicolas Matentzoglu,

HyeongSik Kim, et al. 2023. “Structured Prompt

Interrogation and Recursive Extraction of Semantics

(SPIRES): A Method for Populating Knowledge Bases

Using Zero-Shot Learning,” April. http://arxiv.org

/abs/2304.02711.

Chakraborty, Tanujit, Ujjwal Reddy K S, Shraddha M Naik,

Madhurima Panja, and Bayapureddy Manvitha. 2024.

“Ten Years of Generative Adversarial Nets (GANs): A

Survey of the State-of-the-Art.” Machine Learning:

Science and Technology 5 (1): 011001. https://doi.

org/10.1088/2632-2153/ad1f77.

Danzer, Alex. 2024. “Generative AI, Conversational AI,

Predictive Analytics – What Works Best for Preventive

Healthcare.” MPulse Blogs. mPulse Official Web Site.

2024.

Fan, Shuangfei, and Bert Huang. 2019. “Labeled Graph

Generative Adversarial Networks.” https://arxiv.org/

abs/1906.03220.

Fan, Shuangfei, Virginia Tech, and Bert Huang. 2019.

“Deep Generative Models for Labelled Graphs.” ICLR

Workshop DeepGenStruct, 1–10.

Feng, Yue, Shuchang Liu, Zhenghai Xue, Qingpeng Cai,

Lantao Hu, Peng Jiang, Kun Gai, and Fei Sun. 2023. “A

Large Language Model Enhanced Conversational Re-

commender System.” https://arxiv.org/abs/2308.06212

Gutierrez de Esteban, Nicolas. 2023. “Revolutionizing

Decision-Making with Generative AI: Our Experience

with Ask Grapho.” Intellico Blog. intellico.ai. 2023.

Han, Qiushi, Haitong Chen, Haoxiang Fan, and Zhenhui

Peng. 2023. “ReDBot: Exploring Conversational

Recommendation for Decision-Making Support in

Group Chats.” In Proceedings of the Eleventh

International Symposium of Chinese CHI, 73–80. New

York, USA: ACM. https://doi.org/10.1145/3629606.

3629615.

LLM-Assisted Augmented Intelligence for Context-Aware Decision Support: Current Trends and Integrated Approach

415

Jia, Nan, Xiaolin Tian, Wenxing Gao, and Licheng Jiao.

2023. “Deep Graph-Convolutional Generative

Adversarial Network for Semi-Supervised Learning on

Graphs.” Remote Sensing 15 (12): 3172. https://doi.

org/10.3390/rs15123172.

Jiang, Feifeng, Jun Ma, Christopher John Webster, Alain

J.F. Chiaradia, Yulun Zhou, Zhan Zhao, and Xiaohu

Zhang. 2024. “Generative Urban Design: A Systematic

Review on Problem Formulation, Design Generation,

and Decision-Making.” Progress in Planning 180

(February): 100795. https://doi.org/10.1016/j.progress.

2023.100795.

Kommineni, Vamsi Krishna, Birgitta König-Ries, and

Sheeba Samuel. 2024. “From Human Experts to

Machines: An LLM Supported Approach to Ontology

and Knowledge Graph Construction,” March.

http://arxiv.org/abs/2403.08345.

Kusiak, Andrew. 2025. “Generative Artificial Intelligence

in Smart Manufacturing.” Journal of Intelligent

Manufacturing 36 (August): 1–3. https://doi.org/

10.1007/s10845-024-02480-6.

Lam, Michelle S., Janice Teoh, James A. Landay, Jeffrey

Heer, and Michael S. Bernstein. 2024. “Concept

Induction: Analyzing Unstructured Text with High-

Level Concepts Using LLooM.” In Proceedings of the

CHI Conference on Human Factors in Computing

Systems, 1–28. New York, NY, USA: ACM.

https://doi.org/10.1145/3613904.3642830.

Láng, B., and B. Dömsödi. 2024. “Integration of AI and

Metaheuristics in Educational Software: A Hybrid

Approach to Exercise Generation.” International

Journal of Emerging Technologies in Learning 19 (6):

38–51.

Lawless, Connor, Jakob Schoeffer, Lindy Le, Kael Rowan,

Shilad Sen, Cristina St. Hill, Jina Suh, and Bahareh

Sarrafzadeh. 2024. “‘I Want It That Way’: Enabling

Interactive Decision Support Using Large Language

Models and Constraint Programming.” ACM Transac-

tions on Interactive Intelligent Systems 14 (3): 1–33.

https://doi.org/10.1145/3685053.

Le, Thao. 2023. “Explaining the Uncertainty in AI-Assisted

Decision Making.” Proceedings of the AAAI

Conference on Artificial Intelligence 37 (13): 16119–

20. https://doi.org/10.1609/aaai.v37i13.26920.

Le, Thao, Tim Miller, Liz Sonenberg, and Ronal Singh.

2024. “Towards the New XAI: A Hypothesis-Driven

Approach to Decision Support Using Evidence.”

https://arxiv.org/abs/2402.01292.

Le, Thao, Tim Miller, Ruihan Zhang, Liz Sonenberg, and

Ronal Singh. 2024. “Visual Evaluative AI: A

Hypothesis-Driven Tool with Concept-Based

Explanations and Weight of Evidence.” https://arxiv.

org/abs/2402.01292.

Liao, Wenjie, Xinzheng Lu, Yifan Fei, Yi Gu, and Yuli

Huang. 2024. “Generative AI Design for Building

Structures.” Automation in Construction 157: 105187.

https://doi.org/10.1016/j.autcon.2023.105187.

Lipton, Peter. 1990. “Contrastive Explanation.” Royal

Institute of Philosophy Supplement 27 (March): 247–

66. https://doi.org/10.1017/S1358246100005130.

Liu, Ollie, Deqing Fu, Dani Yogatama, and Willie

Neiswanger. 2024. “DeLLMa: Decision Making Under

Uncertainty with Large Language Models,” February.

http://arxiv.org/abs/2402.02392.

Megawaty, Megawaty, and Maria Ulfa. 2020. “Decision

Support System Methods: A Review.” Journal of

Information Systems and Informatics 2 (1): 192–201.

https://doi.org/10.33557/journalisi.v2i1.63.

Miller, Tim. 2023. “Explainable AI Is Dead, Long Live

Explainable AI!” In

2023 ACM Conf. on Fairness,

Accountability, and Transparency, 333–42. New York,

USA: ACM. https://doi.org/10.1145/3593013.3594001

Neuhaus, Fabian. 2023. “Ontologies in the Era of Large

Language Models – a Perspective.” Applied Ontology

18 (4): 399–407. https://doi.org/10.3233/AO-230072.

Pitkäranta, Tapio, and Leena Pitkäranta. 2024. “Bridging

Human and AI Decision-Making with LLMs: The

RAGADA Approach.” In Proceedings of the 26th

International Conference on Enterprise Information

Systems, 812–19. SCITEPRESS - Science and

Technology Publications. https://doi.org/10.5220/

0012705000003690.

Ramaul, Laavanya, Paavo Ritala, and Mika Ruokonen.

2024. “Creational and Conversational AI Affordances:

How the New Breed of Chatbots Is Revolutionizing

Knowledge Industries.” Business Horizons 67 (5): 615–

27. https://doi.org/10.1016/j.bushor.2024.05.006.

Sahoo, Pranab, Ayush Kumar Singh, Sriparna Saha, Vinija

Jain, Samrat Mondal, and Aman Chadha. 2024. “A

Systematic Survey of Prompt Engineering in Large

Language Models: Techniques and Applications.”

https://arxiv.org/abs/2402.07927.

Simon, Herbert. 1979. “Rational Decision Making in

Business Organizations.” American Economic

Association 69 (4): 493–513.

Smirnov, Alexander, Nikolay Shilov, and Andrew

Ponomarev. 2022. “Context-Aware Knowledge

Management as an Enabler for Human-Machine

Collective Intelligence.” In Knowledge Discovery,

Knowledge Engineering and Knowledge Management.

IC3K 2020. Communications in Computer and

Information Science, Vol 1608, 94–116. Springer,

Cham. https://doi.org/10.1007/978-3-031-14602-2_5.

Tu, Tao, Mike Schaekermann, Anil Palepu, Khaled Saab,

Jan Freyberg, Ryutaro Tanno, Amy Wang, et al. 2025.

“Towards Conversational Diagnostic Artificial

Intelligence.” Nature 642 (8067): 442–50.

https://doi.org/10.1038/s41586-025-08866-7.

Zhao, Xiutian, Ke Wang, and Wei Peng. 2024. “An

Electoral Approach to Diversify LLM-Based Multi-

Agent Collective Decision-Making.” In Proceedings of

the 2024 Conference on Empirical Methods in Natural

Language Processing, 2712–27. Stroudsburg, PA,

USA: Association for Computational Linguistics.

https://doi.org/10.18653/v1/2024.emnlp-main.158.

Zhou, Dawei, Lecheng Zheng, Jiejun Xu, and Jingrui He.

2019. “Misc-GAN: A Multi-Scale Generative Model

for Graphs.” Frontiers in Big Data 2 (April).

https://doi.org/10.3389/fdata.2019.00003.

KMIS 2025 - 17th International Conference on Knowledge Management and Information Systems

416