Improving Multi-label Emotion Classification by Integrating both General and Domain-specific Knowledge

The lack of human-friendly explanations from Artificial Intelligence (AI)—based decisions is a major concern for high stake applications. Towards mitigating the concern, Explainable AI (XAI) is an emerging area of research. In pre-modeling explainability, which is one of the notions of explainability, one can introduce explainability before training the model. One can also introduce explainability during training of the model or after training the model, in general, both of which are known as post-hoc explainability. Unfortunately, post-hoc explainability is not readily transparent and can be misleading, as it explains after the decision has been made and can be optimized to placate a subjective demand which is a form of bias. As a result, the explanation from it can be misleading even though it seems plausible. Explainability should incorporate knowledge from different domains such as philosophy, psychology, and cognitive science, so that the explanation is not just based on the researcher’s intuition of what constitutes a good explanation. Domain knowledge is an abstract, fuzzy, and high-level concept over the problem domain. For instance, in an image classification problem, the domain knowledge could be a dog has four legs, or a zebra has stripes, etc. However, the use of domain knowledge for explainability is under-focused and bound to problem-specific requirements. This chapter focuses on the notion of a pre-modeling explainability of AI-based “black box” models using domain knowledge. We demonstrate the collection and application of domain knowledge along with the quantification of explainability on an intrusion detection problem. Although AI-based Intrusion Detection Systems (IDS) provide accelerated speeds in intrusion detection, the response is still at a human speed when there is a human in the loop. The lack of explainability of an AI-based model is a key reason for this bottleneck as a human analyst has to understand the prediction before making the final decision towards mitigation of a problem. To mitigate this issue, in this chapter, we incorporate the CIA principle (i.e., domain knowledge) in an AI-based black box model for better explainability and generalizability of the model and demonstrate the process in detail. A major portion of this chapter is a compilation of our previously published work [1–4]. We start with a brief discussion of the problem, network intrusion detection and prevention, that we use to demonstrate our approach. Then we briefly introduced relevant domain knowledge and how it can be integrated into the overall architecture. In Sect. 3, we describe our experiments, followed by Sect. 4 which contains a discussion on results from the experiments. We conclude with limitations and future work in Sect. 5.

Online services are interested in solutions to opinion mining, which is the problem of extracting aspects, opinions, and sentiments from text. One method to mine opinions is to leverage the recent success of pre-trained language models which can be fine-tuned to obtain high-quality extractions from reviews. However, fine-tuning language models still requires a non-trivial amount of training data. In this paper, we study the problem of how to significantly reduce the amount of labeled training data required in fine-tuning language models for opinion mining. We describe Snippext, an opinion mining system developed over a language model that is fine-tuned through semi-supervised learning with augmented data. A novelty of Snippext is its clever use of a two-prong approach to achieve state-of-the-art (SOTA) performance with little labeled training data through: (1) data augmentation to automatically generate more labeled training data from existing ones, and (2) a semi-supervised learning technique to leverage the massive amount of unlabeled data in addition to the (limited amount of) labeled data. We show with extensive experiments that Snippext performs comparably and can even exceed previous SOTA results on several opinion mining tasks with only half the training data required. Furthermore, it achieves new SOTA results when all training data are leveraged. By comparison to a baseline pipeline, we found that Snippext extracts significantly more fine-grained opinions which enable new opportunities of downstream applications.

In statistical machine translation, training data usually have the characteristics of diverse sources, multiple themes, different genre, and are often not in accordance with the domain of target text to be translated, resulting in domain adaptive problem. The existing adaptive methods for statistical machine translation aim for the target text and focus on the selection of training data and the adjustment of translation models. These approaches have not specified explicit domain labels for texts or data. This study gives explicit domain labels and uses two examples for specific context knowledge, (1) Domain knowledge based on Chinese Thesaurus are applied to assign domain labels of Chinese Library Classification Number to Chinese texts; (2) Two-dimensional lexicalized domain knowledge, such as Semantic Category and Application Scenarios, is used to label Japanese sentence. Based on the obtained domain labels for development data and test data, the training data can be filtered to achieve the goal of domain consistency. Experiments show that only a part of the training data can gain a comparable translation performance to the whole training data. This shows that the method is efficient and feasible.

Significant progress has been made in sentiment analysis over the past few years, especially due to the application of deep neural language models. However, there is a problem of transferability of trained models from one domain to another, especially for less studied languages such as Russian. We propose an approach to build cross-domain sentiment analysis models based on a two-stage procedure: first, we fine-tune a pre-trained RuBERT language model on a combined non-domain corpus, and then fine-tune this model on a small domain corpus. We conducted large-scale experiments with 30 sentiment annotated corpora across 12 domains. In order to increase the representativeness of news texts with high-quality annotation, we created a novel RuNews corpus, containing 1,823 news articles annotated by sentiment. The results show that fine-tuning the model using a small number (about several hundred) of annotated domain texts can significantly improve the performance of sentiment analysis for a new domain (on average by 4.6 p.p.). We also obtained the state-of-the-art results for 7 out of 14 test corpora.

Structural Health Monitoring (SHM) is a condition-based maintenance technology using sensing systems. In SHM, the use of domain knowledge is essential: it motivates the use of machine learning approaches; it can be used to extract damage sensitive features and interpret the results by machine learning. This work focuses on two SHM problems: damage identification and substructure clustering. Our solutions to address them are based on machine learning techniques and robust feature extraction using domain knowledge. In the first problem, damage sensitive features were extracted using a frequency domain decomposition, followed by a robust one-class support vector machine for damage detection. In the second problem, a novel clustering technique and spectral moment feature were utilised for substructure grouping and anomaly detection. These methods were evaluated using data from lab-based structures and data collected from the Sydney Harbour Bridge. We obtained high damage detection accuracies and were able to assess damage severity. Furthermore, the clustering technique was able to group substructures of similar behaviour and detect spatial anomalies.

Since Computer Science students are used to applying both General Purpose Programming Languages (GPPLs) and Domain-Specific Languages (DSLs), Generative Artificial Intelligence based on Language Models (LMs) can help them on automatic tasks, allowing them to focus on more creative tasks and higher skills. However, the teaching and evaluation of technical tasks in Computer Science can be inefficient and prone to errors. Thus, the main objective of this article is to explore the performance of LMs compared to that of undergraduate Computer Science students in a specific case study: designing and implementing RESTful APIs DSLs. This research aims to determine if LMs can enhance the efficiency and accuracy of these processes. Our case study involved 39 students and 5 different LMs that must use the two DSLs we also designed for their task assignment. To evaluate performance, we applied uniform criteria to student and LMs-generated solutions, enabling a comparative analysis of accuracy and effectiveness. With a case study comparing performance between students and LMs, this article contributes to checking to what extent LMs are able to carry out software development tasks involving the use of new DSLs specially designed for highly specific settings in a similar way as well-qualified Computer Science students are able to. The results underscore the importance of welldefined DSLs and effective prompting processes for optimal LM performance. Specifically, LMs demonstrated high variability in task execution, with two GPT-based LMs achieving similar grades to those scored by the best of the students for every task, obtaining 0.78 and 0.92 on a normalized scale [0, 1], with 0.23 and 0.14 Standard Deviation for ChatGPT-4 and ChatGPT-4o respectively. After the experience, we can conclude that a well-defined DSL and a proper prompting process, providing the LM with metadata, persistent prompts, and a good knowledge base, are crucial for good LM performance. When LMs receive the right prompts, both large and small LMs can achieve excellent results depending on the task.

Learning objects paradigm is widely adopted in e-learning environments. Learning objects management can be improved using semantic technologies from ontology engineering and the semantic web. In this article we use a semantic model of the repository to improve both learning objects retrieval and composition. The use of domain knowledge enables automatic reasoning and makes the system able to import new domain models and use them to interrogate the repository (Arrigoni Neri 2005). Learning objects composition is one of the main challenges in e-learning management systems and can be improved exploiting ontological reasoning. The building of a course can be carried out in two phases, in the first phase we compose concept level entities to obtain an outline of the course, then we fill the outline with actual resources from the repository. Both phases can use ontology-based models to capture specific domain knowledge (Arrigoni Neri 2006). In order to provide an intuitive and expressive ontology representation, we briefly propose a graphical syntax for the well-known ontology web language (OWL).

Purely, data-driven large scale image classification has been achieved using various feature descriptors like SIFT, HOG etc. Major milestone in this regards is Convolutional Neural Networks (CNN) based methods which learn optimal feature descriptors as filters. Little attention has been given to the use of domain knowledge. Ontology plays an important role in learning to categorize images into abstract classes where there may not be a clear visual connect between category and image, for example identifying image mood — happy, sad and neutral. Our algorithm combines CNN and ontology priors to infer abstract patterns in Indian Monument Images. We use a transfer learning based approach in which, knowledge of domain is transferred to CNN while training (top down transfer) and inference is made using CNN prediction and ontology tree/priors (bottom up transfer ). We classify images to categories like Tomb, Fort and Mosque. We demonstrate that our method improves remarkably over logistic classifier and other transfer learning approach. We conclude with a remark on possible applications of the model and note about scaling this to bigger ontology.

Deep language models, like ELMo, BERT and GPT, have achieved impressive results on several natural language tasks. These models are pretrained on large corpora of unlabeled general domain text and later supervisedly trained on downstream tasks. An optional step consists of finetuning the language model on a large intradomain corpus of unlabeled text, before training it on the final task. This aspect is not well explored in the current literature. In this work, we investigate the impact of this step on named entity recognition (NER) for Portuguese legal documents. We explore different scenarios considering two deep language architectures (ELMo and BERT), four unlabeled corpora and three legal NER tasks for the Portuguese language. Experimental findings show a significant improvement on performance due to language model finetuning on intradomain text. We also evaluate the finetuned models on two general-domain NER tasks, in order to understand whether the aforementioned improvements were really due to domain similarity or simply due to more training data. The achieved results also indicate that finetuning on a legal domain corpus hurts performance on the general-domain NER tasks. Additionally, our BERT model, finetuned on a legal corpus, significantly improves on the state-of-the-art performance on the LeNER-Br corpus, a Portuguese language NER corpus for the legal domain.

Twitter has emerged as a critical platform for capturing public sentiment, offering a valuable source for sentiment analysis. This study presents a comparative evaluation of two BERT (Bidirectional Encoder Representations from Transformers) models—baseline and fine-tuned—targeted at analyzing Indonesian-language tweets. Employing the CRISP-DM framework, the methodology encompasses automated data crawling, comprehensive text pre-processing (including case folding, cleaning, tokenisation, normalisation, and data augmentation), and model development using the IndoBERT-base-p1 architecture. The experimental results reveal that the fine-tuned BERT model achieves significantly improved performance over the non-optimized model, with accuracy, precision, recall, and F1-score values reaching 91%, 0.91, 0.90, and 0.91, respectively. These findings indicate the fine-tuned model's superior ability to capture linguistic subtleties and contextual sentiment features within informal social media text. Furthermore, the model is deployed in a web-based application for real-time sentiment classification, demonstrating its practical applicability. This study underscores the effectiveness of Fine-tuning in enhancing BERT-based sentiment analysis for low-resource languages. It highlights its potential for informing decisionmaking in digital communication, marketing, and policy research contexts.

Energy management in handheld devices is becoming a daunting task with the growing number of accelerators, increasing memory demands and high computing capacities required to support applications with stringent QoS needs. Current DVFS techniques that modulate power states of a single hardware component, or even recent proposals that manage multiple components, can lose out opportunities for attaining high energy efficiencies that may be possible by leveraging application domain knowledge. Thus, this paper proposes a coordinated multi-component energy optimization mechanism for handheld devices, where the energy profile of different components such as CPU, memory, GPU and IP cores are considered in unison to trigger the appropriate DVFS state by exploiting the application domain knowledge. Specifically, we show that for the important class of frame-based applications, the domain knowledge - frame processing rates, component utilization and available slack - can be used to decide effective DVFS states for each component from among the numerous choices. With such knowledge, rather than a brute force search of all speed setting choices, we propose two simpler heuristics, called Greedy policy and Kaldor-Hicks compensation policy, to make the decisions at frame boundaries. Our evaluations with 7 commonly-used Android apps show that our domain-aware coordinated DVFS policies have 23% better energy efficiency than the conventionally used Android governors, and are within ~9% of an optimal policy that does not drop any frames.

There are currently few methods that can be applied to malware classification problems which don't require domain knowledge to apply. In this work, we develop our new SHWeL feature vector representation, by extending the recently proposed Lempel-Ziv Jaccard Distance. These SHWeL vectors improve upon LZJD's accuracy, outperform byte n-grams, and allow us to build efficient algorithms for both training (a weakness of byte n-grams) and inference (a weakness of LZJD). Furthermore, our new SHWeL method also allows us to directly tackle the class imbalance problem, which is common for malware-related tasks. Compared to existing methods like SMOTE, SHWeL provides significantly improved accuracy while reducing algorithmic complexity to O(N). Because our approach is developed without the use of domain knowledge, it can be easily re-applied to any new domain where there is a need to classify byte sequences.

The increasing complexity of modern aircraft creates a need for a larger number of caution and warning devices. But more alerts require more memorization and higher work loads for the pilot and tend to induce a higher probability of errors. Therefore, we have developed an architecture for a flight expert system (FLES) to assist pilots in monitoring, diagnosing and recovering from in-flight faults. A prototype of FLES has been implemented. A sensor simulation model was developed and employed to provide FLES with the airplane status information during the diagnostic process. The simulator is based partly on the Lockheed Advanced Concept System (ACS), a future generation airplane, and partly on the Boeing 737, an existing airplane. A distinction between two types of faults, maladjustments and malfunctions, has led us to take two approaches to fault diagnosis. These approaches are evident in two FLES subsystems: the flight phase monitor and the sensor interrupt handler. The specific problem addressed in these subsystems has been that of integrating information received from multiple sensors with domain knowledge in order to assess abnormal situations during airplane flight. This paper describes our reasons for handling malfunctions and maladjustments separately and the use of domain knowledge in the diagnosis of each.

Knowledge-based region labeling for remote sensing image interpretation

During the design optimization of the excavator boom, there are many design variables and complicated processes. The original optimization methods mainly focused on the optimization of mathematical models, and they lacked consideration in the use of domain knowledge, design-specification knowledge, expert experience knowledge, and historical examples. In order to comprehensively utilize the domain knowledge and expert experience knowledge, this study uses the optimization process analysis, uses knowledge expression and coding processing technology to encode the boom structure, builds an optimal design coding system based on knowledge guidance, and realizes the automatic optimization design of the boom structure. In the process of constructing the knowledge-oriented optimization system, to realize the reuse of the knowledge of the boom structure design in the numerical optimization iteration, a knowledge processing flowchart of the boom structure design is constructed. The concept of “shape distance” is proposed to judge the similarity feature matrix of the boom structure coding. To evaluate whether the stress distribution is uniform, a fast prediction model based on stress characteristic regions is constructed. The research results show that, under the comprehensive consideration of the four working conditions, the knowledge-guided optimization of the boom structure can avoid the deformity in the optimization process, accelerate the calculation speed of the optimization model, and improve the optimization quality of the model.