Diana Inkpen
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Diana Inkpen has a total of 83 co-authors. Here are the top co-authors by number of shared publications:
Aline Villavicencio
8 shared publications- Branimir K. Boguraev
IBM Thomas J. Watson Research Center USA
8 shared publications - Ruslan Mitkov
University of Wolverhampton
8 shared publications Sebastian Padó
8 shared publicationsSandra Elfiky
7 shared publicationsJing Jiang
6 shared publications
Latest Publications
- 1 May 2026
- Published by Springer Nature
- p. 314-326
A Variance-Invariance Approach to Better Word Embeddings
- 1 May 2026
- Published by Springer Nature
- p. 314-326
A Variance-Invariance Approach to Better Word Embeddings
- 1 May 2026
- Published by Springer Nature
- p. 387-402
When One Size Doesn’t Fit All: A Study on Intrinsic Iterative Self-correction in LLMs for Hallucination Control
- 1 May 2026
- Published by Springer Nature
- p. 387-402
When One Size Doesn’t Fit All: A Study on Intrinsic Iterative Self-correction in LLMs for Hallucination Control
- 26 February 2026
- Published by Wiley in Computational Intelligence
- Vol. 42 (2)
Forty Years of Computational Intelligence : A Bibliometric Retrospective
In 2025, the Computational Intelligence journal celebrates the 40th anniversary. Motivated by this event, this paper conducts a bibliometric analysis of the journal between 1985 and 2024 by using the Scopus and Web of Science databases. The aim is to explore the research trends, development, topic clusters, and key contributors in Computational Intelligence . The study also employs the VOSviewer and bibliometrix software to visualize co‐citation networks, bibliographic coupling, and co‐occurrence patterns to map the intellectual structure of the journal. The analysis shows that early theories were introduced primarily by Canadian and American researchers, while more recent contributions are increasingly coming from developing countries such as China and India, reflecting the global changing research trends. The exploration reveals the evolution of topics over time, highlighting the transition from early symbolic artificial intelligence and logic‐based reasoning towards modern developments in deep learning, big data, and machine learning. The study highlights Computational Intelligence 's past achievements while outlining its future direction, emphasizing collaboration, diversity, and innovation as key to sustaining its academic excellence.
- 26 February 2026
- Published by Wiley in Computational Intelligence
- Vol. 42 (2)
Forty Years of Computational Intelligence : A Bibliometric Retrospective
In 2025, the Computational Intelligence journal celebrates the 40th anniversary. Motivated by this event, this paper conducts a bibliometric analysis of the journal between 1985 and 2024 by using the Scopus and Web of Science databases. The aim is to explore the research trends, development, topic clusters, and key contributors in Computational Intelligence . The study also employs the VOSviewer and bibliometrix software to visualize co‐citation networks, bibliographic coupling, and co‐occurrence patterns to map the intellectual structure of the journal. The analysis shows that early theories were introduced primarily by Canadian and American researchers, while more recent contributions are increasingly coming from developing countries such as China and India, reflecting the global changing research trends. The exploration reveals the evolution of topics over time, highlighting the transition from early symbolic artificial intelligence and logic‐based reasoning towards modern developments in deep learning, big data, and machine learning. The study highlights Computational Intelligence 's past achievements while outlining its future direction, emphasizing collaboration, diversity, and innovation as key to sustaining its academic excellence.
- Open Access
- 19 May 2025
- Published by PubPub
Enhancing Legal Text Entailment: Evaluating Model Architectures, Training Approaches, and Interpretability
This study investigates the effectiveness of various model architectures and training strategies for legal text classification, with a focus on entailment classification for case decisions from the Federal Court of Canada. We compare the performance of RoBERTa models with and without domain-specific further pretraining, to larger language models such as Llama 2, Llama 3, and GPT-4o adapted to the task by using prompt engineering and LORA fine-tuning. Additionally, we investigate different methods that can be used to explain the decisions of the models and evaluate their adequacy, understandability, trustworthiness, and sufficiency. Our findings suggest that for legal entailment classification, domain-specific pretraining can improve performance for smaller models, while larger language models show promise in outperforming prompt engineering for classification when fine-tuned with LoRA, as well as in generating more interpretable explanations. To the best of our knowledge, this is the first study in the context of Canadian legal AI to explore the effects of further pretraining on small and large language models, and the integration of language model adaptation and explainability into one system for legal text entailment classification.
- 19 May 2025
- Published by PubPub
Open AccessEnhancing Legal Text Entailment: Evaluating Model Architectures, Training Approaches, and Interpretability
This study investigates the effectiveness of various model architectures and training strategies for legal text classification, with a focus on entailment classification for case decisions from the Federal Court of Canada. We compare the performance of RoBERTa models with and without domain-specific further pretraining, to larger language models such as Llama 2, Llama 3, and GPT-4o adapted to the task by using prompt engineering and LORA fine-tuning. Additionally, we investigate different methods that can be used to explain the decisions of the models and evaluate their adequacy, understandability, trustworthiness, and sufficiency. Our findings suggest that for legal entailment classification, domain-specific pretraining can improve performance for smaller models, while larger language models show promise in outperforming prompt engineering for classification when fine-tuned with LoRA, as well as in generating more interpretable explanations. To the best of our knowledge, this is the first study in the context of Canadian legal AI to explore the effects of further pretraining on small and large language models, and the integration of language model adaptation and explainability into one system for legal text entailment classification.
- Open Access
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 5201-5215
A Context-Aware Contrastive Learning Framework for Hateful Meme Detection and Segmentation
Xuanyu Su, Yansong Li, Diana Inkpen, Nathalie Japkowicz. Findings of the Association for Computational Linguistics: NAACL 2025. 2025.
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 5201-5215
Open AccessA Context-Aware Contrastive Learning Framework for Hateful Meme Detection and Segmentation
Xuanyu Su, Yansong Li, Diana Inkpen, Nathalie Japkowicz. Findings of the Association for Computational Linguistics: NAACL 2025. 2025.
- Open Access
- 1 November 2024
- Published by Agora University of Oradea in International Journal of Computers Communications & Control
- Vol. 19 (6)
Evaluating and Mitigating Gender Bias in Generative Large Language Models
The examination of gender bias, alongside other demographic biases like race, nationality, and religion, within generative large language models (LLMs), is increasingly capturing the attention of both the scientific community and industry stakeholders. These biases often affect generative LLMs, influencing popular products and potentially compromising user experiences. A growing body of research is dedicated to enhancing gender representations in natural language processing (NLP) across a spectrum of generative LLMs. This paper explores the current research focused on identifying and evaluating gender bias in generative LLMs. A comprehensive investigation is conducted to evaluate and mitigate gender bias across five distinct generative LLMs. The mitigation strategies implemented yield significant improvements in gender bias scores, with performance enhancements of up to 46% compared to zero-shot text generation approaches. Additionally, we explore how different levels of LLM precision and quantization impact gender bias, providing insights into how technical factors influence bias mitigation strategies. By tackling these challenges and suggesting areas for future research, we aim to contribute to the ongoing discussion about gender bias in language technologies, promoting more equitable and inclusive NLP systems.
- 1 November 2024
- Published by Agora University of Oradea in International Journal of Computers Communications & Control
- Vol. 19 (6)
Open AccessEvaluating and Mitigating Gender Bias in Generative Large Language Models
The examination of gender bias, alongside other demographic biases like race, nationality, and religion, within generative large language models (LLMs), is increasingly capturing the attention of both the scientific community and industry stakeholders. These biases often affect generative LLMs, influencing popular products and potentially compromising user experiences. A growing body of research is dedicated to enhancing gender representations in natural language processing (NLP) across a spectrum of generative LLMs. This paper explores the current research focused on identifying and evaluating gender bias in generative LLMs. A comprehensive investigation is conducted to evaluate and mitigate gender bias across five distinct generative LLMs. The mitigation strategies implemented yield significant improvements in gender bias scores, with performance enhancements of up to 46% compared to zero-shot text generation approaches. Additionally, we explore how different levels of LLM precision and quantization impact gender bias, providing insights into how technical factors influence bias mitigation strategies. By tackling these challenges and suggesting areas for future research, we aim to contribute to the ongoing discussion about gender bias in language technologies, promoting more equitable and inclusive NLP systems.
- Open Access
uOttawa at LegalLens-2024: Transformer-based Classification Experiments
This paper presents the methods used for LegalLens-2024 shared task, which focused on detecting legal violations within unstructured textual data and associating these violations with potentially affected individuals. The shared task included two subtasks: A) Legal Named Entity Recognition (L-NER) and B) Legal Natural Language Inference (L-NLI). For subtask A, we utilized the spaCy library, while for subtask B, we employed a combined model incorporating RoBERTa and CNN. Our results were 86.3% in the L-NER subtask and 88.25% in the L-NLI subtask. Overall, our paper demonstrates the effectiveness of transformer models in addressing complex tasks in the legal domain. The source code for our implementation is publicly available at https://github.com/NimaMeghdadi/uOttawa-at-LegalLens-2024-Transformer-based-Classification
Open AccessuOttawa at LegalLens-2024: Transformer-based Classification Experiments
This paper presents the methods used for LegalLens-2024 shared task, which focused on detecting legal violations within unstructured textual data and associating these violations with potentially affected individuals. The shared task included two subtasks: A) Legal Named Entity Recognition (L-NER) and B) Legal Natural Language Inference (L-NLI). For subtask A, we utilized the spaCy library, while for subtask B, we employed a combined model incorporating RoBERTa and CNN. Our results were 86.3% in the L-NER subtask and 88.25% in the L-NLI subtask. Overall, our paper demonstrates the effectiveness of transformer models in addressing complex tasks in the legal domain. The source code for our implementation is publicly available at https://github.com/NimaMeghdadi/uOttawa-at-LegalLens-2024-Transformer-based-Classification
- Open Access
- 25 June 2024
- Published by Faculty of Organisation and Informatics in Journal of information and organizational sciences
- Vol. 48 (1) , 1-17
Relevancy between Anchor Text and Wikipedia. A Web Search Framework
The overall volume of data available on the Internet is growing rapidly while finding relevant documents is becoming increasingly difficult. Moreover, queries entered by users are unique, unstructured and often ambiguous while the process has changed dramatically from standard query languages that governed by strict syntax rules to unstructured strings. In Web information retrieval, search paradigms used term occurrences to weight document content prior to any boosting stage. PageRank algorithm, for instance, was used integrated techniques to enhance post retrieval document relevancy to adequately compromise the overall process in two stages. Nevertheless, hypertexts in Web have been used for improving the quality of search results for the most common type of queries. Our main premise is that hypertexts play an important role for ranking documents in IR such as margining between user queries and consensus hypertext. We propose a new algorithm that uses term impact technology for compromising hypertext weighting in Web along with Wikipedia for efficiently find most relevant documents among large set of results. Our experimental results showed that Wikipedia could efficiently improve document relevancy rank when combined with hypertexts for exhibit robust and very good short-term process capability.
- 25 June 2024
- Published by Faculty of Organisation and Informatics in Journal of information and organizational sciences
- Vol. 48 (1) , 1-17
Open AccessRelevancy between Anchor Text and Wikipedia. A Web Search Framework
The overall volume of data available on the Internet is growing rapidly while finding relevant documents is becoming increasingly difficult. Moreover, queries entered by users are unique, unstructured and often ambiguous while the process has changed dramatically from standard query languages that governed by strict syntax rules to unstructured strings. In Web information retrieval, search paradigms used term occurrences to weight document content prior to any boosting stage. PageRank algorithm, for instance, was used integrated techniques to enhance post retrieval document relevancy to adequately compromise the overall process in two stages. Nevertheless, hypertexts in Web have been used for improving the quality of search results for the most common type of queries. Our main premise is that hypertexts play an important role for ranking documents in IR such as margining between user queries and consensus hypertext. We propose a new algorithm that uses term impact technology for compromising hypertext weighting in Web along with Wikipedia for efficiently find most relevant documents among large set of results. Our experimental results showed that Wikipedia could efficiently improve document relevancy rank when combined with hypertexts for exhibit robust and very good short-term process capability.
- Open Access
- 27 May 2024
- Published by PubPub
Instruction Tuning of LLMs for Multi-label EmotionClassification in Social Media Content
In this research, we use Instruction Tuning (IT) prompts to enhance the controllability of generative Large Language Models (LLMs) in order to facilitate multi-label emotion classification of user tweets. We use instruction tags to represent the data records as {(tweet, question, answer)} prompts which are then used to fine-tune LLMs in a supervised fashion. Multi-label emotion classification is a challenging research area having widespread applications. For training and evaluation, we use the Semeval-2018 multi-label emotion classification dataset (Task E-c) that is labeled with 11 emotion labels (one or more labels are assigned to each tweet). We perform fine-tuning of a pre-trained GPT2-model with different prompt formats and then use the fine-tuned model to perform inference on the test records to predict the answer in the {(tweet, question, answer)} prompt, as next-word generation task. We compare the performance improvements over the current best-performing systems and GPT 3.5 in a zero-shot setting. Our proposed method and prompt format managed to surpass the accuracy of the current state-of-the-art system and shows that the performance of instruction tuning of lighter GPT2-based models is better in performance than a larger GPT 3.5 in zero-shot implementation.
- 27 May 2024
- Published by PubPub
Open AccessInstruction Tuning of LLMs for Multi-label EmotionClassification in Social Media Content
In this research, we use Instruction Tuning (IT) prompts to enhance the controllability of generative Large Language Models (LLMs) in order to facilitate multi-label emotion classification of user tweets. We use instruction tags to represent the data records as {(tweet, question, answer)} prompts which are then used to fine-tune LLMs in a supervised fashion. Multi-label emotion classification is a challenging research area having widespread applications. For training and evaluation, we use the Semeval-2018 multi-label emotion classification dataset (Task E-c) that is labeled with 11 emotion labels (one or more labels are assigned to each tweet). We perform fine-tuning of a pre-trained GPT2-model with different prompt formats and then use the fine-tuned model to perform inference on the test records to predict the answer in the {(tweet, question, answer)} prompt, as next-word generation task. We compare the performance improvements over the current best-performing systems and GPT 3.5 in a zero-shot setting. Our proposed method and prompt format managed to surpass the accuracy of the current state-of-the-art system and shows that the performance of instruction tuning of lighter GPT2-based models is better in performance than a larger GPT 3.5 in zero-shot implementation.
- Open Access
- 27 May 2024
- Published by PubPub
Explainable Prompt-based Approaches for Sentiment Analysis of Movie Reviews
This study investigates the application of prompting and prompt learning with large language models for sentiment analysis on IMDB movie reviews, achieving a notable 98.53\% accuracy in sentiment classification through prompt learning with the Llama 2 model. The research highlights the effectiveness of fine-tuning language models on specific prompts, demonstrating enhanced performance across various models. Additionally, it pioneers the application of XAI techniques to prompt learning systems, offering human-understandable explanations for model predictions and evaluating them with metrics like sufficiency, justification, understandability, and trustworthiness. The work presents a novel framework for sentiment analysis and a comprehensive examination of XAI methods in prompt learning, contributing valuable insights into model behavior and the significance of prompt engineering.
- 27 May 2024
- Published by PubPub
Open AccessExplainable Prompt-based Approaches for Sentiment Analysis of Movie Reviews
This study investigates the application of prompting and prompt learning with large language models for sentiment analysis on IMDB movie reviews, achieving a notable 98.53\% accuracy in sentiment classification through prompt learning with the Llama 2 model. The research highlights the effectiveness of fine-tuning language models on specific prompts, demonstrating enhanced performance across various models. Additionally, it pioneers the application of XAI techniques to prompt learning systems, offering human-understandable explanations for model predictions and evaluating them with metrics like sufficiency, justification, understandability, and trustworthiness. The work presents a novel framework for sentiment analysis and a comprehensive examination of XAI methods in prompt learning, contributing valuable insights into model behavior and the significance of prompt engineering.