Reasoning over large-scale knowledge graphs has long been dominated by embedding-based methods, which focus on representing entities and relationships in vector spaces to perform inference tasks. Despite advancements in knowledge graph completion (KGC), challenges such as data sparsity and the lack of interpretability persist. These issues are critical in biomedical areas, where prediction accuracy and explainability directly impact decision-making.

In light of these limitations, this paper proposes a KGC approach that allows it to leverage both globally structured patterns and locally analogical information. Aiming to enhance the reasoning capability and completeness of knowledge graphs, we develop a better joint inference approach based on the rules mining and knowledge graph embedding. Specifically, our approach employs knowledge graph embedding techniques to discover analogically similar triples, which are then used to construct more relevant and accurate inference paths. Subsequently, rule mining is integrated to extract structured knowledge patterns through analogical reasoning.

Experimental results demonstrate that our approach performs better than the traditional Knowledge Graph Embedding Model in some specified KGC tasks, which depend highly on the inner construction of the dataset and its relation types. The multi-hop nature of DRKG aligns well with rule-based approaches, where learned rules generalize over multiple instances, aiding in missing link prediction. Beyond improved performance, our method enhances explainability, facilitating transparent inference and enabling backtracking of key prediction results—an important feature for biomedical and high-stakes AI applications.

This thesis investigates the enhancement of sentence decomposition in Large Language Models (LLMs) through the integration of linguistic features, including constituency parsing, dependency parsing, and abstract meaning representation. Traditional decomposition methods, which often rely on rule-based approaches, struggle with highly intricate sentences. By incorporating detailed linguistic features and employing reasoning steps and supervision in prompts, we aim to improve the comprehension and decomposition capabilities of LLMs. Experimental results show that integrating these linguistic features has the potential to improve decomposition performance. However, these enhancements also introduce new error types and increase computational costs. This study identifies several error types in the programs generated by LLMs, including format, decomposition, and conversion errors, emphasizing the need for further refinement in model training and prompt design. This research lays a step toward more accurate and efficient processing of complex sentences in LLMs, encouraging ongoing development and optimization in this field.

Language models (LLMs) have demonstrated impressive performance on knowledge-intensive tasks like question answering when supported by external knowledge. However, their success relies not only on their reasoning capabilities and the accuracy of the external knowledge but also on the truthfulness of the prompts provided. False premises in prompts can lead to "hallucinations," where the generated content appears plausible but is factually incorrect. This issue is common in online questions, particularly when users search for information on unfamiliar topics, leading to confirmation bias in information retrieval.

Existing methods for detecting hallucinations may not effectively handle false premises, as they can be misled by coherent responses that align with the false premises. Fact-checking methods may also be unsuitable, as LLMs can exhibit sycophantic behavior in attempting to satisfy user requirements.

To address this challenge, we propose a False Premise Detection with Abductive Reasoning FPDAR method for question answering with LLMs. Abductive reasoning enables backward thinking, minimizing less plausible assumptions and working towards the correct answer through a bottom-up approach. FPDAR is designed as a plug-and-play module that can be integrated after the question-answering process.

FPDAR employs a two-stage abductive reasoning process. First, it infers the most plausible question intent based on factual context and generated response without considering the potentially problematic question as input. This allows for the identification of false premises by comparing the inferred intent with the original question. Second, abductive reasoning helps generate a more plausible explanation aligned with factual context, increasing the likelihood of correctness and ruling out less plausible, potentially hallucinated responses.

To the best of our knowledge, this is the first study introducing abductive reasoning for identifying and diagnosing false premises. We conduct extensive experiments on two question-answering benchmarks containing false premises to validate the effectiveness of FPDAR. The results show that FPDAR can achieve high accuracy in terms of response correctness, although it may struggle to effectively detect false premises. Nevertheless, it achieves substantial accuracy improvements over state-of-the-art methods.

Trust is a fundamental component in human-AI relationships, serving as a critical element of user acceptance and satisfaction, particularly within the realm of Decision Support Systems (DSS). The technological advances in conversational user interfaces (CUIs) such as ChatGPT and digital assistants (e.g., Alexa) allow laypeople to interact with DSS without knowing the mechanisms behind them.
Extensive research has explored the benefits of CUIs and strives to improve their usability and adoption rate. However, while interacting with such CUIs, how to facilitate proper user trust for decision support is still under-explored. To address the research gap, we aim to test the impact of emotional expressiveness in CUIs on building user trust.

To analyze the impact of emotional expressiveness in CUIs to build user trust and whether voice-based CUI is more efficient in building user trust compared to text-based CUI. We implemented a conversational interface with varying emotional expressiveness that can serve six conditions: two text-based and four voice-based. Text-based CUIs are differentiated by lexical expressiveness. Voice-based CUIs are varying in both lexical expressiveness and prosodic expressiveness. Regardless of the modality and emotional expressiveness, each CUI serves as an interactive medium for users with the DSS, which supports them to find a suitable house given a scenario.

Through an empirical study (N = 151), the experimental results are insufficient to conclude the impact of prosodic expressiveness and lexical expressiveness on user trust and usability in CUIs. In addition, we did not find any statistically significant difference between text-based and voice-based CUIs in trust or perceived usability.

Our findings can potentially be explained by the uncanniness effect [46]: initially, increased emotional expressiveness in a chatbot could positively influence user trust, but over time this could turn into a negative impact. These results offer a potential way to explain the complex dynamics of trust in conversational DSS and some implications in chatbot design within the context of DSS. Our findings
can benefit the future design and development of conversational agents-based DSS by considering emotional expressiveness.

Commonsense knowledge based question answer- ing is a recent topic that has seen a surge in inter- est. Yet most models obtain general data, this pa- per looks at obtaining query-specific similar con- cepts using first and second-order proximity to- gether with BERT-based retrieval. Using these query-specific concepts new commonsense knowl- edge can be obtained using a Game with a pur- pose. Results show that this current implementa- tion leaves room for improvement.

Search engines operate as an oracle between user queries and information access: the user types the input and receives back the information requested. To accomplish the task, search engines need to interpret human language and, most importantly, comprehend the underlying user intents of a query. With this process, they can retrieve the most appropriate sources of information. The purpose of our research is to introduce a new,
hierarchical taxonomy that better depicts the underlying intents of users asking questions online (on search engines and Q&A platforms). Throughout our study, we first review the prior work and findings on the topic. We assemble a new dataset with queries aggregated from MS Marco, AskReddit and Quora. We examine its questions and label them to construct a new fine-grained ontology. Our examination continues with the integration of
Deep Learning (DL) models and Active Learning (AL) to evaluate the quality of our work. The results show that the taxonomy can effectively assess users’ goals. Our taxonomy, the dataset composed and the codebase are publicly available to support future research.

Commonsense knowledge is a type of knowledge consisting of facts that humans use every day. Humans make queries in search engines with different user intents, and some of them can be answered by knowledge tuples. Different types of knowledge are stored differently in the knowledge bases. Being aware of the types of commonsense knowledge required to answer the queries can accelerate the process of finding corresponding knowledge for the search engines to give a response to users. For some queries with specific user intents, it is not possible to be answered solely with commonsense knowledge because some analysis and judgment from humans are needed. On the other hand, some queries can be answered with commonsense knowledge tuples and the user intents can have a strong indication of what the knowledge type is required to answer. The research is to look into how to map queries and their user intents to knowledge types and explore the impacts of user intents in the knowledge type classification. There was no existing dataset that had annotations on both user intents and knowledge types. In this work, the described dataset was created. Observations of the created dataset and experiments on three classifiers with accuracy being around 0.99 were conducted. The results show that user intents generally help the classification of the type of commonsense knowledge.

Commonsense knowledge plays a key role in human intelligence. It is knowledge possessed by most humans that helps them in everyday situations. One possible way is to store the knowledge in four types. Each piece is either positive or negative, and generative or discriminative. For efficient retrieval and storage, a uniform model is needed. Existing models for commonsense knowledge are not fit for negative and discriminative knowledge. The aim of this paper is to create a uniform model to store both positive and negative generative and discriminative knowledge tuples. Models are evaluated on a set of generalized queries as well as on the storage they require. Four possible models were evaluated of which two were the most promising: the generative model and the combined model. The generative model is efficient in storage and retrieving generative knowledge for concepts, but relatively slow in distinguishing concepts. Combining the generative model with discriminative tuples gives the combined model, a model that is the most efficient for all queries but expensive in storage. Which of the two models is most suitable depends on the application and the available resources.

Common sense is knowledge that most humans have, but machines do not. Generally, computer knowledge bases make use of positive (known) knowledge. However, in addition to positive common sense knowledge, there is also negative. Negative knowledge represent facts that are known to be untrue, like "a cat does not have fins". This knowledge is important in order for a machine to make assumptions the same way a human does. This research proposes ways to combine and organize the positive and negative knowledge tuples in a unified way. The two main ways that are looked into are table-like and graph-like organizations. These are analyzed based on different requirements and important queries are defined. Based on the requirements a recommendation is made. The research also takes a look at a solution that makes use of complex number space and suggest how this solution could be further researched and improved in the future.

Commonsense knowledge (CK) in artificial intelligence (AI), is an expanding field of research. Because CK is intrinsically implicit, current datadriven machine learning models are still far from competent compared to humans in commonsense reasoning tasks. To minimize the gap between machine learning models with the goal of artificial general intelligence (AGI), researchers propose to collect such CK from human and text corpus for commonsense augmentation. Over the years there have been many ways that commonsense knowledge (in AI) has been implemented. However, there has not been a systematic review conducted on how CK can be organized, structured and categorized. The aim of this paper is to do a literature survey on how existing knowledge sources organize, structure and categorize within the general frameworks of CK. The organization can decide the design schema of a knowledge graph (KG), the structuring decides the format and the categorization decides what dimensions and criteria a KG employs.

Commonsense knowledge is information that all humans own and use to interpret common situations and react to them accordingly. This kind of information is necessary for the training of artificial intelligence models to reach a performance as close as possible to human performance. Researchers have developed methods that use crowdsourcing to collect this kind of knowledge from the general public. This research focuses on systematically surveying the existing literature about these methods. We created a taxonomy to describe and compare the existing work based on the following three measures that were the most common ones reported: efficiency, cost, and quality.

Common sense knowledge (CSK) comes naturally to humans, but is very hard for computers to comprehend. However it is critical for machines to behave intelligently, and as such collecting CSK has become a prevalent field of research. Whilst a lot of research has been done to develop CSK acquisition methods, not much work has been done to survey the literature that already exists. Furthermore the surveys that have been done are outdated, and as such there is a clear gap in the literature. This paper will survey the different approaches to CSK acquisition and evaluate their effectiveness, as a way of gauging their real life applicability. It will also compare the current \textit{state of the art} methods, to some previous work to illustrate the progress that has been made and project that into the future. Furthermore this paper will also create a taxonomy categorizing the surveyed literature in order give a better overview of existing methods. Finally, from the literature surveyed it is clear that these methods have made a lot of progress, but aren't quite yet at the same level as human performance. Nevertheless they have become robust enough to be deployed in real applications.

Commonsense knowledge is the key of human intelligence in generalizing their knowledge to deal with complex tasks. Over the past years, a lot of research has been done in both natural language processing (NLP) and computer vision (CV) on leveraging commonsense knowledge to improve AI models. However, no systematic comparisons of existing work have been made between the two domains. Therefore this survey aims to provide an overview of how commonsense knowledge is used within NLP and CV and how research varies between these two domains and what future challenges it may hold. An observation made from this survey is that leveraging commonsense is more difficult in CV than NLP, as commonsense is mostly incorporated textually and datasets need to be filtered to make them more relevant for visual commonsense. We hope to promote further research and create a better understanding of commonsense knowledge and its applications with this survey.

Artificial Intelligence (AI) is increasingly helping people with all kinds of tasks, due to its promising capabilities. In some tasks, an AI system by itself will take over tasks, but in other tasks, an AI system making decisions on its own would be undesired due to ethical and legal reasons. In those cases, AI can still be of help by forming human-AI teams, in which humans get advice from the AI system helping them with making their final decisions. Human-AI teams are for instance used in the medical and legal fields. One problem arises, in which instances should one trust an AI system and in which not? Trusting the AI system when it is correct and trusting yourself when you are correct, results in a high appropriate reliance. If users appropriately rely on AI systems, it is possible to achieve complementary team performance, which is better than any single teammate. However, as known from previous literature, people struggle with assessing their performance and knowing how well they perform compared to peers. When one overestimates their performance this can be because of a dual burden, due to the lack of skill they also lack the skill to accurately estimate their performance. This phenomenon is called the Dunning-Kruger Effect (DKE). This raises questions about whether the inability to estimate their own capabilities would also reflect on their assessment of the AI system its performance.

In this thesis we look at how the DKE affects (appropriate) reliance on AI systems and if so, how such effects due to the DKE can be mitigated. The effects of the DKE and possible mitigation are being tested through an empirical study (N = 249). The attempt at mitigation is done by including a tutorial intervention, which has been proved in previous research to be useful in decreasing the DKE. The tutorial intervention is aimed at revealing the weaknesses of the participant and making them aware of their miscalibrated self-estimation. Furthermore, in this thesis, the effects of revealing how the AI system makes its decisions through explainable AI (XAI) are explored. The XAI consisted of highlights from logic unit-based explanations, it should allow participants to gain more understanding of the AI advice. This thesis shows how this will affect user self-assessment and reliance on the AI system.

We found that participants who overestimate themselves tend to rely less on the AI system, compared to participants that had an accurate or underestimation of their performance. After the tutorial participants have a better calibration of their self-assessment. While the designed tutorial intervention can help participants calibrate their self-assessment, it fails to promote (appropriate) reliance. Furthermore, the logic units-based explanations did not improve accurate self-assessing or increase user (appropriate) reliance on AI systems.
This thesis shows the importance of considering cognitive biases when dealing with human-AI teams and invites more research on how to handle and mitigate the DKE in human-AI decision making.