Capturing expertise and enabling efficient information retrieval are critical in the energy sector, where high staff turnover can lead to significant knowledge loss. Retrieval Augmented Generation (RAG) offers a solution by grounding Large Language Model (LLM) outputs in documented sources, but its effectiveness is limited by reliance on general-purpose embeddings. We present Wikatoni, an agentic AI system for energy engineering workflows that integrates a novel domain-specific embedding model. Wikatoni combines fine-tuned embeddings with agentic RAG, metadata filtering, and hybrid retrieval to improve document search, automated reporting, and workflow efficiency. Evaluation on internal enterprise offshore energy data shows that the domain-adapted embedding improves recall by 10%, and Wikatoni agentic RAG further increases answer accuracy by 14% compared to vanilla RAG with the base embedding model, achieving the best overall performance in context recall, faithfulness, and answer accuracy.

Automated Essay Scoring (AES) presents a key opportunity to improve student experience while reducing the administrative burden of academic staff. However existing methods for AES are reliant on large volumes of data and fail to consider the ordinal aspect of grading. As a result, when an institution introduces a new assessment, there may be no data available to train algorithms. In this paper, we demonstrate that metric learning architectures, specifically Prototypical Networks, offer robust performance on few-shot ordinal classification essay grading tasks. We introduce three novel weighted prototype calculation strategies designed to enhance class representation in ordinal few-shot text classification. These strategies improve how class knowledge is modeled from limited examples by refining the way prototypes are computed, incorporating weighted mechanisms for better differentiation. Results across four datasets show that our methods outperform existing baselines and the current state-of-the-art in ordinal few-shot text classification. Additionally, we compare our approach with three large language models (LLMs) using a prompt-based approach to few-shot learning and find that we achieve superior or comparable performance in all evaluated tasks.

Explainable AI (XAI) can greatly enhance user trust and satisfaction in AI-assisted decision-making processes. Recent findings suggest that a single explainer may not meet the diverse needs of multiple users in an AI system; indeed, even individual users may require multiple explanations. This highlights the necessity for a “multi-shot” approach, employing a combination of explainers to form what we introduce as an “explanation strategy”. Tailored to a specific user or a user group, an “explanation experience” describes interactions with personalised strategies designed to enhance their AI decision-making processes. The iSee platform is designed for the intelligent sharing and reuse of explanation experiences, using Case-based Reasoning to advance best practices in XAI. The platform provides tools that enable AI system designers, i.e. design users, to design and iteratively revise the most suitable explanation strategy for their AI system to satisfy end-user needs. All knowledge generated within the iSee platform is formalised by the iSee ontology for interoperability. We use a summative mixed methods study protocol to evaluate the usability and utility of the iSEE platform with six design users across varying levels of AI and XAI expertise. Our findings confirm that the iSee platform effectively generalises across applications and its potential to promote the adoption of XAI best practices.

Regina Ofori-Boateng, Magaly Aceves-Martins, Nirmalie Wiratunga, Carlos Moreno-Garcia. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Systematic Review (SR) are foundational to influencing policies and decision-making in healthcare and beyond. SRs thoroughly synthesise primary research on a specific topic while maintaining reproducibility and transparency. However, the rigorous nature of SRs introduces two main challenges: significant time involved and the continuously growing literature, resulting in potential data omission, making most SRs become outmoded even before they are published. As a solution, AI techniques have been leveraged to simplify the SR process, especially the abstract screening phase. Active learning (AL) has emerged as a preferred method among these AI techniques, allowing interactive learning through human input. Several AL software have been proposed for abstract screening. Despite its prowess, how the various parameters involved in AL influence the software’s efficacy is still unclear. This research seeks to demystify this by exploring how different AL strategies, such as initial training set, query strategies etc. impact SR automation. Experimental evaluations were conducted on five complex medical SR datasets, and the GLM model was used to interpret the findings statistically. Some AL variables, such as the feature extractor, initial training size, and classifiers, showed notable observations and practical conclusions were drawn within the context of SR and beyond where AL is deployed. • This study explores optimal Active Learning (AL) combinations for systematic reviews (SRs). • Smaller initial training samples improve performance metrics in datasets. • TF-IDF consistently outperformed Doc2Vec and S-BERT. • Certainty and Uncertainty strategies gave comparative results and effectively interacted with the TF-IDF. • The impact of AL variables in SR automation varies according to the specific dataset.

Explainable Artificial Intelligence (XAI) is an emerging field within Artificial Intelligence (AI) that has provided many methods that enable humans to understand and interpret the outcomes of AI systems. However, deciding on the best explanation approach for a given AI problem is currently a challenging decision-making task. This paper presents the iSee project, which aims to address some of the XAI challenges by providing a unifying platform where personalized explanation experiences are generated using Case-Based Reasoning. An explanation experience includes the proposed solution to a particular explainability problem and its corresponding evaluation, provided by the end user. The ultimate goal is to provide an open catalog of explanation experiences that can be transferred to other scenarios where trustworthy AI is required.

Counterfactual Explanations (cf-XAI) describe the smallest changes in feature values necessary to change an outcome from one class to another. However, many cf-XAI methods neglect the feasibility of those changes. In this paper, we introduce a novel approach for presenting cf-XAI in natural language (Natural-XAI), giving careful consideration to actionable and comprehensible aspects while remaining cognizant of immutability and ethical concerns. We present three contributions to this endeavor. Firstly, through a user study, we identify two types of themes present in cf-XAI composed by humans: content-related, focusing on how features and their values are included from both the counterfactual and the query perspectives; and structure-related, focusing on the structure and terminology used for describing necessary value changes. Secondly, we introduce a feature actionability taxonomy with four clearly defined categories, to streamline the explanation presentation process. Using insights from the user study and our taxonomy, we created a generalisable template-based natural language generation (NLG) method compatible with existing explainers like DICE, NICE, and DisCERN, to produce counterfactuals that address the aforementioned limitations of existing approaches. Finally, we conducted a second user study to assess the performance of our taxonomy-guided NLG templates on three domains. Our findings show that the taxonomy-guided Natural-XAI approach (n-XAIT) received higher user ratings across all dimensions, with significantly improved results in the majority of the domains assessed for articulation, acceptability, feasibility, and sensitivity dimensions.

The need to understand the decision-making mechanisms of deep learning networks has led to a growing effort in exploring both modal-dependent and model-agnostic research methods. Although both of these ideas provide transparency for automated decision making, most methodologies focus on either using the modal-gradients (model- dependent) or ignoring the model internal states and reasoning with a model's behavior/outcome (model-agnostic) to instances. In this work, we propose a unified explanation approach that given an instance combines both model-dependent and agnostic explanations to produce an explanation set. The generated explanations are not only consistent in the neighborhood of a sample but can highlight causal relationships between image content and the outcome. We use the Wireless Capsule Endoscopy (WCE) domain to illustrate the effectiveness of our explanations. The saliency maps generated by our approach are competitive on the softmax information score.

Counterfactual explanations highlight actionable knowledge which helps to understand how a machine learning model outcome could be altered to a more favourable outcome. Understanding actionable corrections in source code analysis can be critical to proactively mitigate security attacks that are caused by known vulnerabilities. In this paper, we present the DisCERN explainer for discovering counterfactuals for code vulnerability correction. Given a vulnerable code segment, DisCERN finds counterfactual (i.e. non-vulnerable) code segments and recommends actionable corrections. DisCERN uses feature attribution knowledge to identify potentially vulnerable code statements. Subsequently, it applies a substitution-focused correction, suggesting suitable fixes by analysing the nearest-unlike neighbour. Overall, DisCERN aims to identify vulnerabilities and correct them while preserving both the code syntax and the original functionality of the code. A user study evaluated the utility of counterfactuals for vulnerability detection and correction compared to more commonly used feature attribution explainers. The study revealed that counterfactuals foster positive shifts in mental models, effectively guiding users towards making vulnerability corrections. Furthermore, counterfactuals significantly reduced the cognitive load when detecting and correcting vulnerabilities in complex code segments. Despite these benefits, the user study showed that feature attribution explanations are still more widely accepted than counterfactuals, possibly due to the greater familiarity with the former and the novelty of the latter. These findings encourage further research and development into counterfactual explanations, as they demonstrate the potential for acceptability over time among developers as a reliable resource for both coding and training.

Counterfactual explanations focus on "actionable knowledge" to help end-users understand how a machine learning outcome could be changed to a more desirable outcome. For this purpose a counterfactual explainer needs to discover input dependencies that relate to outcome changes. Identifying the minimum subset of feature changes needed to action an output change in the decision is an interesting challenge for counterfactual explainers. The DisCERN algorithm introduced in this paper is a case-based counter-factual explainer. Here counterfactuals are formed by replacing feature values from a nearest unlike neighbour (NUN) until an actionable change is observed. We show how widely adopted feature relevance-based explainers (i.e. LIME, SHAP), can inform DisCERN to identify the minimum subset of "actionable features". We demonstrate our DisCERN algorithm on five datasets in a comparative study with the widely used optimisation-based counterfactual approach DiCE. Our results demonstrate that DisCERN outperformed DiCE by minimising both the number of feature changes and the amount of change necessary to create good counterfactual explanations.

The general aim of a cognitive Cyber Physical System of Systems (CPSoS) is to provide managed access to data in a smart fashion such that sensing and actuation capabilities are connected. Whilst there is significant funding and research devoted to this area, focus remains purely on creating bespoke systems. This paper presents a novel approach, based on a set of components to leverage Situational Awareness and Smart Actuation in large manufacturing industries with the focus on enabling predictive maintenance for asset and abnormal situation management. This paper presents a novel generic platform, named AtiCoS, that combines case-based and common-sense reasoning, as the enabling methodologies for enhancing CPSoS with cognitive capabilities.

Abstract Theoretical frameworks in psychology map the relationships between emotions and sentiments. In this paper, we study the role of such mapping for computational emotion detection from text (e.g., social media) with an aim to understand the usefulness of an emotion‐rich corpus of documents (e.g., tweets) to learn polarity lexicons for sentiment analysis. We propose two different methods that leverage a corpus of emotion‐labelled tweets to learn word‐polarity lexicons. The proposed methods model the emotion corpus using a generative unigram mixture model, combined with the emotion‐sentiment mapping proposed in psychology for automated generation of word‐polarity lexicons that capture emotion‐rich vocabulary. We comparatively evaluate the quality of the proposed mixture model in learning emotion‐aware sentiment lexicons with those generated using supervised latent dirichlet allocation (sLDA) and word‐document‐frequency (WDF) statistics. Sentiment analysis experiments on benchmark Twitter data sets confirm the quality of our proposed lexicons. Further, a comparative analysis with sLDA, WDF‐based emotion‐aware lexicons, and standard sentiment lexicons that are agnostic to emotion knowledge suggests that the proposed lexicons lead to a significantly better performance in both sentiment classification and sentiment intensity prediction tasks.

Triplet networks are deep metric learners which learn to optimise a feature space using similarity knowledge gained from training on triplets of data simultaneously. The architecture relies on the triplet loss function to optimise its weights based upon the distance between triplet members. Composition of input triplets therefore directly impacts the quality of the learned representations, meaning that a training scheme which optimises their formation is crucial. However, an exhaustive search for the best triplets is prohibitive unless the search for triplets is confined to smaller training regions or batches. Accordingly, current triplet mining approaches use informed selection applied only to a random minibatch, but the resulting view fails to exploit areas of complexity in the feature space. In this work, we introduce a locality-sensitive batching strategy, which uses the locality of examples to create batches as an alternative to the commonly adopted randomly minibatching. Our results demonstrate this method to offer better performance on three image and two text classification tasks with statistical significance. Importantly most of these gains are incrementally realised with as little as 25% of the training iterations.

Exercise adherence is a key component of digital behaviour change interventions for the self-management of musculoskeletal pain. Automated monitoring of exercise adherence requires sensors that can capture patients performing exercises and Machine Learning (ML) algorithms that can recognise exercises. In contrast to ambulatory activities that are recognisable with a wrist accelerometer data; exercises require multiple sensor modalities because of the complexity of movements and the settings involved. Exercise Recognition (ExR) pose many challenges to ML researchers due to the heterogeneity of the sensor modalities (e.g. image/video streams, wearables, pressure mats). We recently published MEx, a benchmark dataset for ExR, to promote the study of new and transferable HAR methods to improve ExR and benchmarked the state-of-the-art ML algorithms on 4 modalities. The results highlighted the need for fusion methods that unite the individual strengths of modalities. In this paper, we explore fusion methods with a focus on attention and propose a novel multi-modal hybrid attention fusion architecture mHAF for ExR. We achieve the best performance of 96.24% (F1-measure) with a modality combination of a pressure mat, a depth camera and an accelerometer on the thigh. mHAF significantly outperforms multiple baselines and the contribution of architecture components are verified with an ablation study. The benefits of attention fusion are clearly demonstrated by visualising attention weights; showing how mHAF learns feature importance and modality combinations suited for different exercise classes. We highlight the importance of improving deployability and minimising obtrusiveness by exploring the best performing 2 and 3 modality combinations.

Sentiment lexicon is a crucial resource for opinion mining from social media content. However, standard off-the-shelve lexicons are static and typically do not adapt, in content and context, to a target domain. This limitation, adversely affects the effectiveness of sentiment analysis algorithms. In this paper, we introduce the idea of distant-supervision to learn a domain-focused lexicon to improve coverage and sentiment context of terms. We present a weighted strategy to integrate scores from the domain-focused with the static lexicon to generate a hybrid lexicon. Evaluations of this hybrid lexicon on social media text show superior sentiment classification over either of the individual lexicons. A further comparative study with typical machine learning approaches to sentiment analysis also confirms this position. We also present promising results from our investigations into the transferability of this distant-supervised hybrid lexicon on three different social media.

We consider the problem of segmenting text documents that have a two-part structure such as a problem part and a solution part. Documents of this genre include incident reports that typically involve description of events relating to a problem followed by those pertaining to the solution that was tried. Segmenting such documents into the component two parts would render them usable in knowledge reuse frameworks such as Case-Based Reasoning. This segmentation problem presents a hard case for traditional text segmentation due to the lexical inter-relatedness of the segments. We develop a two-part segmentation technique that can harness a corpus of similar documents to model the behavior of the two segments and their inter-relatedness using language models and translation models respectively. In particular, we use separate language models for the problem and solution segment types, whereas the inter-relatedness between segment types is modeled using an IBM Model 1 translation model. We model documents as being generated starting from the problem part that comprises of words sampled from the problem language model, followed by the solution part whose words are sampled either from the solution language model or from a translation model conditioned on the words already chosen in the problem part. We show, through an extensive set of experiments on real-world data, that our approach outperforms the state-of-the-art text segmentation algorithms in the accuracy of segmentation, and that such improved accuracy translates well to improved usability in Case-based Reasoning systems. We also analyze the robustness of our technique to varying amounts and types of noise and empirically illustrate that our technique is quite noise tolerant, and degrades gracefully with increasing amounts of noise.

In image retrieval, most existing approaches that incorporate local features produce high dimensional vectors, which lead to a high computational and data storage cost. Moreover, when it comes to the retrieval of generic real-life images, randomly generated patches are often more discriminant than the ones produced by corner/blob detectors. In order to tackle these problems, we propose a novel method incorporating local features with a hybrid sampling (a combination of detector-based and random sampling). We take three large data collections for the evaluation: MIRFlickr, ImageCLEF, and a collection from British National Geological Survey. The overall performance of the proposed approach is better than the performance of global features and comparable with the current state-of-the-art methods in content-based image retrieval. One of the advantages of our method when compared with others is its easy implementation and low computational cost. Another is that hybrid sampling can improve the performance of other methods based on the ``bag of visual words'' approach.

Textual reuse is an integral part of textual case-based reasoning (TCBR) which deals with solving new problems by reusing previous similar problem-solving experiences documented as text. We investigate the role of text reuse for text authoring applications that involve feedback or review generation. Generally providing feedback in the form of assigning a rating from a likert scale is far easier compared to articulating explanatory feedback as text. When previous feedback generated about the same or similar objects are maintained as cases, there is opportunity for knowledge reuse. In this paper, we show how compositional and transformational adaptation techniques can be applied once sentences in a given case are aligned to relevant structured attribute values. Three text reuse algorithms are introduced and evaluated on a dataset gathered from online Hotel reviews from TripAdvisor. Here cases consists of both structured sub-rating attributes together with textual feedback. Generally, aligned sentences linked to similar sub-rating values are clustered together and prototypical sentences are then extracted to enable reuse across similar authors. Experiments show a close similarity between our proposed texts and actual human edited review text. We also found that problems with variability in vocabulary are best addressed when prototypes are formulated from larger sets of similar sentences in contrast to smaller sets from local neighbourhoods.