Intention Detection Research Articles

Bridging Human and Machine Intelligence: Reverse-Engineering Radiologist Intentions for Clinical Trust and Adoption

In the rapidly evolving landscape of medical imaging, the integration of artificial intelligence (AI) with clinical expertise offers unprecedented opportunities to enhance diagnostic precision and accuracy. Yet, the "black box" nature of AI models often limits their integration into clinical practice, where transparency and interpretability are important. This paper presents a novel system leveraging the Large Multimodal Model (LMM) to bridge the gap between AI predictions and the cognitive processes of radiologists. This system consists of two core modules, Temporally Grounded Intention Detection (TGID) and Region Extraction (RE). The TGID module predicts the radiologist's intentions by analyzing eye gaze fixation heatmap videos and corresponding radiology reports. Additionally, the RE module extracts regions of interest that align with these intentions, mirroring the radiologist’s diagnostic focus. This approach introduces a new task, radiologist intention detection, and is the first application of Dense Video Captioning (DVC) in the medical domain. By making AI systems more interpretable and aligned with radiologist’s cognitive processes, this proposed system aims to enhance trust, improve diagnostic accuracy, and support medical education. Additionally, it holds the potential for automated error correction, guiding junior radiologists, and fostering more effective training and feedback mechanisms. This work sets a precedent for future research in AI-driven healthcare, offering a pathway towards transparent, trustworthy, and human-centered AI systems. We evaluated this model using NLG(Natural Language Generation), time-related, and vision-based metrics, demonstrating superior performance in generating temporally grounded intentions on REFLACX and EGD-CXR datasets. This model also demonstrated strong predictive accuracy in overlap scores for medical abnormalities and effective region extraction with high IoU(Intersection over Union), especially in complex cases like cardiomegaly and edema. These results highlight the system's potential to enhance diagnostic accuracy and support continuous learning in radiology. We are also releasing the source code for our project, available here.

Few-shot intent detection with mutual information and contrastive learning

Few-shot intent detection is a challenging task. Most existing methods only focus on acquisition of generalization knowledge in known classes, or on the adaptation situation of meta-learning tasks. However, these methods lack both the learning of beneficial information for intent detection and the learning of discriminative representation of intent keywords. Solving these problems can effectively improve the model’s ability to recognize text intent and thus effectively execute user commands. Therefore, in this paper, we propose a few-shot intent detection framework based on Mutual Information and Contrastive Learning (MICL). It improves the robustness of the model learning of intent keywords through mutual information maximization, thus improving the accuracy of intent detection. Additionally, maximization of mutual information enhances the intent representation consistency on two text views,11In this paper, the “two text views” of the text represent the original text and its corresponding data augmentation text.which helps to promote the efficiency of sample contrastive learning. Specifically, we use a neural network to approximately estimate the joint probability distribution and marginal probability distribution of two text views and maximize the mutual information through back-propagation. In order to enable the model to focus on intent keywords in text, we introduce a sample-level contrastive learning. It helps the model to learn discriminative representations of intent keywords by comparing the similarity of text from two text views. In addition, to prevent overfitting of the model on known classes, we introduce a class-level contrastive learning. We regard it as regularization of known classes, aiming to improve the model generalization performance on unknown classes. Our method achieves state-of-the-art performance on four public datasets.22Our method code is available: https://github.com/YS19999/MICL.

A Segment Augmentation and Prediction Consistency Framework for Multi-Label Unknown Intent Detection

Multi-label unknown intent detection is a challenging task where each utterance may contain not only multiple known but also unknown intents. To tackle this challenge, pioneers proposed to predict the intent number of the utterance first, then compare it with the results of known intent matching to decide whether the utterence contains unknown intent(s). Though they have made remarkable progress on this task, their methods still suffer from two important issues: (1) It is inadequate to extract multiple intents using only utterance encoding; (2) Optimizing two sub-tasks (intent number prediction and known intent matching) independently leads to inconsistent predictions. In this article, we propose to incorporate segment augmentation rather than only use utterance encoding to better detect multiple intents. We also design a prediction consistency module to bridge the gap between the two sub-tasks. Empirical results on MultiWOZ2.3 and MixSNIPS datasets show that our method achieves state-of-the-art performance and significantly improves the best baseline.

Joint intent detection and slot filling with syntactic and semantic features using multichannel CNN-BiLSTM

Understanding spoken language is crucial for conversational agents, with intent detection and slot filling being the primary tasks in natural language understanding (NLU). Enhancing the NLU tasks can lead to an accurate and efficient virtual assistant thereby reducing the need for human intervention and expanding their applicability in other domains. Traditionally, these tasks have been addressed individually, but recent studies have highlighted their interconnection, suggesting better results when solved together. Recent advances in natural language processing have shown that pretrained word embeddings can enhance text representation and improve the generalization capabilities of models. However, the challenge of poor generalization in joint learning models for intent detection and slot filling remains due to limited annotated datasets. Additionally, traditional models face difficulties in capturing both the semantic and syntactic nuances of language, which are vital for accurate intent detection and slot filling. This study proposes a hybridized text representation method using a multichannel convolutional neural network with three embedding channels: non-contextual embeddings for semantic information, part-of-speech (POS) tag embeddings for syntactic features, and contextual embeddings for deeper contextual understanding. Specifically, we utilized word2vec for non-contextual embeddings, one-hot vectors for POS tags, and bidirectional encoder representations from transformers (BERT) for contextual embeddings. These embeddings are processed through a convolutional layer and a shared bidirectional long short-term memory (BiLSTM) network, followed by two softmax functions for intent detection and slot filling. Experiments on the air travel information system (ATIS) and SNIPS datasets demonstrated that our model significantly outperformed the baseline models, achieving an intent accuracy of 97.90% and slot filling F1-score of 98.86% on the ATIS dataset, and an intent accuracy of 98.88% and slot filling F1-score of 97.07% on the SNIPS dataset. These results highlight the effectiveness of our proposed approach in advancing dialogue systems, and paving the way for more accurate and efficient natural language understanding in real-world applications.

Combining replay and LoRA for continual learning in natural language understanding

Large language models have significantly improved dialogue systems through enhanced capabilities in understanding queries and generating responses. Despite these enhancements, task-oriented dialogue systems—which power many intelligent assistants—face challenges when adapting to new domains and applications. This challenge arises from a phenomenon known as catastrophic forgetting, where models forget previously acquired knowledge when learning new tasks. This paper addresses this issue through continual learning techniques to preserve previously learned knowledge while seamlessly integrating new tasks and domains. We propose Experience Replay Informative-Low Rank Adaptation or ERI-LoRA, a hybrid continual learning method for natural language understanding in dialogue systems that effectively combines the replay-based methods with parameter-efficient techniques. Our experiments on intent detection and slot-filling tasks demonstrate that ERI-LoRA significantly outperforms competitive baselines in continual learning. The results of our catastrophic forgetting experiments demonstrate that ERI-LoRA maintains robust memory stability in the model, demonstrating its effectiveness in mitigating these effects.

A Multimodal Bracelet to Acquire Muscular Activity and Gyroscopic Data to Study Sensor Fusion for Intent Detection.

Researchers have attempted to control robotic hands and prostheses through biosignals but could not match the human hand. Surface electromyography records electrical muscle activity using non-invasive electrodes and has been the primary method in most studies. While surface electromyography-based hand motion decoding shows promise, it has not yet met the requirements for reliable use. Combining different sensing modalities has been shown to improve hand gesture classification accuracy. This work introduces a multimodal bracelet that integrates a 24-channel force myography system with six commercial surface electromyography sensors, each containing a six-axis inertial measurement unit. The device's functionality was tested by acquiring muscular activity with the proposed device from five participants performing five different gestures in a random order. A random forest model was then used to classify the performed gestures from the acquired signal. The results confirmed the device's functionality, making it suitable to study sensor fusion for intent detection in future studies. The results showed that combining all modalities yielded the highest classification accuracies across all participants, reaching 92.3±2.6% on average, effectively reducing misclassifications by 37% and 22% compared to using surface electromyography and force myography individually as input signals, respectively. This demonstrates the potential benefits of sensor fusion for more robust and accurate hand gesture classification and paves the way for advanced control of robotic and prosthetic hands.

ScamGen: Unveiling psychological patterns in tele-scam through advanced template-augmented corpus generation

Telephone scams, with their profound psychological impact, often compel victims to make hasty and severe decisions. Studying these scams is challenging due to the scarcity of comprehensive datasets, a result of the private nature of telephone interactions. In this paper, we introduce ScamGen, a template-based data augmentation technique designed to enhance Chinese telephone scam data. ScamGen leverages psychological insights to generate diverse and realistic scam scenarios, focusing on the psychological dynamics between scammers and victims. This novel approach integrates psychological theory with data augmentation, diverging from traditional methods by emphasizing scammer–victim interactions. Our method begins with a multi-source data collection framework, compiling an initial seed dataset of tele-scam samples. Using sentence- and word-level perturbations, we expand this seed data to create a comprehensive and diverse dataset covering a wide range of scam scenarios. Rigorous evaluations demonstrate that ScamGen outperforms large language models in generating high-quality, varied datasets. Additionally, we develop five deep learning models for intent detection on this dataset, with BERT achieving the highest precision at 86.68%. The dataset, which will be made publicly available, marks a significant step toward understanding scammer tactics and improving tele-scam detection systems.

A linguistics-based approach to refining automatic intent detection in conversational agent design

In this paper, we propose Automatic Intent Detector (AID), a framework for automatic intent detection to facilitate the creation of a conversational agent. AID follows an eight-step process incorporating best practices from the current literature and introducing innovative approaches in certain steps. The most notable innovation within AID is the automatic labeling of clusters, which is based on detailed and sophisticated rules derived from linguistics. These rules focus on morphosyntactic analysis, while also taking into account an aspect of semantic role theory. Furthermore, as for the overall validation of the results obtained, it provides an approach based on the concepts of semantic coherence, variability, and label appropriateness. After describing AID at the technical level, we illustrate the experiments we conducted both on a dataset widely used as benchmark in the literature and on a real corporate dataset. Finally, we present a critical discussion on the results obtained.

Force Myography for Motion Intention Detection Based on 3D-Printed Piezoelectric Sensors

Force Myography for Motion Intention Detection Based on 3D-Printed Piezoelectric Sensors

Intent detection for task‐oriented conversational agents: A comparative study of recurrent neural networks and transformer models

AbstractConversational assistants (CAs) and Task‐oriented ones, in particular, are designed to interact with users in a natural language manner, assisting them in completing specific tasks or providing relevant information. These systems employ advanced natural language understanding (NLU) and dialogue management techniques to comprehend user inputs, infer their intentions, and generate appropriate responses or actions. Over time, the CAs have gradually diversified to today touch various fields such as e‐commerce, healthcare, tourism, fashion, travel, and many other sectors. NLU is fundamental in the natural language processing (NLP) field. Identifying user intents from natural language utterances is a sub‐task of NLU that is crucial for conversational systems. The diversity in user utterances makes intent detection (ID) even a challenging problem. Recently, with the emergence of Deep Neural Networks. New State of the Art (SOA) results have been achieved for different NLP tasks. Recurrent neural networks (RNNs) and Transformer architectures are two major players in those improvements. RNNs have significantly contributed to sequence modelling across various application areas. Conversely, Transformer models represent a newer architecture leveraging attention mechanisms, extensive training data sets, and computational power. This review paper begins with a detailed exploration of RNN and Transformer models. Subsequently, it conducts a comparative analysis of their performance in intent recognition for Task‐oriented (CAs). Finally, it concludes by addressing the main challenges and outlining future research directions.

Enhancing conversational ai with the Rasa framework: intent understanding and NLU pipeline optimization

Implementing the Rasa NLU pipeline allowed intent detection and entity recognition, particularly in complex scenarios with multi-intent queries. Communication within the Rasa NLU pipeline was effectively managed, ensuring seamless data flow between components, which preserved context and enhanced interpretability. The voice assistant developed with STT and TTS capabilities demonstrated robust real-time natural language processing, handling spoken queries efficiently. This confirmed the practical viability of using the Rasa framework for scalable and customizable conversational AI applications. Discussing: The findings underscore the robustness of the Rasa NLU pipeline in handling diverse conversational demands and the flexibility of its components to adapt to different linguistic contexts. The research discusses the potential of integrating sophisticated NLU techniques to create more intuitive and responsive conversational agents, highlighting the critical role of context-aware processing in improving user interaction with AI systems.

A novel model based on a transformer for intent detection and slot filling

Building task-oriented dialogue systems has become a topic of interest in the research community and industry. The task-oriented dialogue system is a closed-domain dialogue system that can perform specific tasks for users. The natural language understanding module of a task-oriented dialogue system is crucial because it is related to a task-oriented dialogue system that provides correctional services for users. The natural language understanding module of a task-oriented dialogue system performs two tasks: intent detection and slot filling. The intent detection task can be regarded as a text classification task; a classification model is trained to predict the intention of the user from the user’s input information. The slot filling task can be regarded as a sequence analysis task; a sequence analysis model is trained to predict the details of the user’s intention. In this paper, we proposed a novel model based on a transformer encoder for intent detection and slot filling. It follows the encoder-decoder structure, including a vanilla Transformer encoder, a bidirectional LSTM encoder, a linear classification decoder for intent detection, and a conditional random field decoder for slot filling. The experimental results on two public datasets show that our proposed model outperforms the existing methods based on the Transformer and can be combined with BERT to achieve better intent detection and slot filling results.

Single-trial movement intention detection estimation in patients with Parkinson’s disease: a movement-related cortical potential study

Objectives. Parkinson patients often suffer from motor impairments such as tremor and freezing of movement that can be difficult to treat. To unfreeze movement, it has been suggested to provide sensory stimuli. To avoid constant stimulation, episodes with freezing of movement needs to be detected which is a challenge. This can potentially be obtained using a brain–computer interface (BCI) based on movement-related cortical potentials (MRCPs) that are observed in association with the intention to move. The objective in this study was to detect MRCPs from single-trial EEG. Approach. Nine Parkinson patients executed 100 wrist movements and 100 ankle movements while continuous EEG and EMG were recorded. The experiment was repeated in two sessions on separate days. Using temporal, spectral and template matching features, a random forest (RF), linear discriminant analysis, and k-nearest neighbours (kNN) classifier were constructed in offline analysis to discriminate between epochs containing movement-related or idle brain activity to provide an estimation of the performance of a BCI. Three classification scenarios were tested: 1) within-session (using training and testing data from the same session and participant), between-session (using data from the same participant from session one for training and session two for testing), and across-participant (using data from all participants except one for training and testing on the remaining participant). Main results. The within-session classification scenario was associated with the highest classification accuracies which were in the range of 88%–89% with a similar performance across sessions. The performance dropped to 69%–75% and 70%–75% for the between-session and across-participant classification scenario, respectively. The highest classification accuracies were obtained for the RF and kNN classifiers. Significance. The results indicate that it is possible to detect movement intentions in individuals with Parkinson’s disease such that they can operate a BCI which may control the delivery of sensory stimuli to unfreeze movement.

EduChatbot: Implementing educational Chatbot for assisting the teaching-learning process by NLP-based hybrid heuristic adopted deep learning framework

Purpose A new Chatbot system is implemented to provide both voice-based and textual-based communication to address student queries without any delay. Initially, the input texts are gathered from the chat and then the gathered text is fed to pre-processing techniques like tokenization, stemming of words and removal of stop words. Then, the pre-processed data are given to the Natural Learning Process (NLP) for extracting the features, where the XLnet and Bidirectional Encoder Representations from Transformers (BERT) are utilized to extract the features. From these extracted features, the target-based fused feature pools are obtained. Then, the intent detection is carried out to extract the answers related to the user queries via Enhanced 1D-Convolutional Neural Networks with Long Short Term Memory (E1DCNN-LSTM) where the parameters are optimized using Position Averaging of Binary Emperor Penguin Optimizer with Colony Predation Algorithm (PA-BEPOCPA). Finally, the answers are extracted based on the intent of a particular student’s teaching materials like video, image or text. The implementation results are analyzed through different recently developed Chatbot detection models to validate the effectiveness of the newly developed model.Design/methodology/approach A smart model for the NLP is developed to help education-related institutions for an easy way of interaction between students and teachers with high prediction of accurate data for the given query. This research work aims to design a new educational Chatbot to assist the teaching-learning process with the NLP. The input data are gathered from the user through chats and given to the pre-processing stage, where tokenization, steaming of words and removal of stop words are used. The output data from the pre-processing stage is given to the feature extraction phase where XLnet and BERT are used. In this feature extraction, the optimal features are extracted using hybrid PA-BEPOCPA to maximize the correlation coefficient. The features from XLnet and features from BERT were given to target-based features fused pool to produce optimal features. Here, the best features are optimally selected using developed PA-BEPOCPA for maximizing the correlation among coefficients. The output of selected features is given to E1DCNN-LSTM for implementation of educational Chatbot with high accuracy and precision.Findings The investigation result shows that the implemented model achieves maximum accuracy of 57% more than Bidirectional long short-term memory (BiLSTM), 58% more than One Dimansional Convolutional Neural Network (1DCNN), 59% more than LSTM and 62% more than Ensemble for the given dataset.Originality/value The prediction accuracy was high in this proposed deep learning-based educational Chatbot system when compared with various baseline works.

Developing a tablet-based brain-computer interface and robotic prototype for upper limb rehabilitation.

The current study explores the integration of a motor imagery (MI)-based BCI system with robotic rehabilitation designed for upper limb function recovery in stroke patients. We developed a tablet deployable BCI control of the virtual iTbot for ease of use. Twelve right-handed healthy adults participated in this study, which involved a novel BCI training approach incorporating tactile vibration stimulation during MI tasks. The experiment utilized EEG signals captured via a gel-free cap, processed through various stages including signal verification, training, and testing. The training involved MI tasks with concurrent vibrotactile stimulation, utilizing common spatial pattern (CSP) training and linear discriminant analysis (LDA) for signal classification. The testing stage introduced a real-time feedback system and a virtual game environment where participants controlled a virtual iTbot robot. Results showed varying accuracies in motor intention detection across participants, with an average true positive rate of 63.33% in classifying MI signals. The study highlights the potential of MI-based BCI in robotic rehabilitation, particularly in terms of engagement and personalization. The findings underscore the feasibility of BCI technology in rehabilitation and its potential use for stroke survivors with upper limb dysfunctions.

A Review on the Form and Complexity of Human–Robot Interaction in the Evolution of Autonomous Surgery

As robotics and intelligence increasingly integrate into surgery, the pivotal role of human–robot interaction (HRI) in surgical procedures and outcomes becomes evident. However, debate rages over whether increasing robot autonomy will result in less human involvement. Some scholars assert that autonomy will reduce human participation, whereas others contend it will result in more complex interactions. To reveal the role of HRI in the evolution of autonomous surgery, this review systematically explores the HRI of robotic surgery with various levels of autonomy. The HRI is examined from both robotic science and clinical practice perspectives, incorporating relevant case studies. Two key components, intention detection and situation awareness, are especially concerned with a brief description of the interfaces and control strategies they rely on. Additional insights are drawn from analogous technologies in aviation, industrial robotics, and autonomous vehicles. The analysis suggests that HRI complexity tends to increase as the robot transitions from no autonomy to conditional autonomy and is predicted to subsequently decrease with a substantial shift in the interaction form when moving toward full autonomy. It is concluded by highlighting challenges from technical and clinical perspectives and delineating research trends in this rapidly evolving field.

Decomposing Task-Relevant Information From Surface Electromyogram for User-Generic Dexterous Finger Force Decoding.

Existing electromyographic (EMG) based motor intent detection algorithms are typically user-specific, and a generic model that can quickly adapt to new users is highly desirable. However, establishing such a model remains a challenge due to high inter-person variability and external interference with EMG signals. In this study, we present a feature disentanglement approach, implemented by an autoencoder-like architecture, designed to decompose user-invariant, motor-task-sensitive high-level representations from user-sensitive, task-irrelevant representations in EMG amplitude features. Our method is user-generic and can be applied to unseen users for continuous multi-finger force predictions. We evaluated our approach on eight subjects, predicting the force of three fingers (index, middle, and ring-pinky) concurrently. We assessed the decoder's performance through a rigorous leave-one-subject-out validation. Our developed approach consistently outperformed both the conventional EMG amplitude method and a commonly used feature projection approach, principal component analysis (PCA), with a lower force prediction error (RMSE: 6.91 ± 0.45 % MVC; R2: 0.835 ± 0.026) and a higher finger classification accuracy (83.0 ± 4.5%). The comparison with the state-of-the-art neural networks further demonstrated the superior performance of our method in user-generic force predictions. Overall, our methods provide novel insights into the development of user-generic and accurate neural decoding for myoelectric control of assistive robotic hands.

Leveraging Intent Detection and Generative AI for Enhanced Customer Support

Customer support plays a pivotal role in shaping customer satisfaction and fostering loyalty within any business. This paper delves into how the integration of intent detection and generative AI (GenAI) can transform customer support systems. At the core of this transformation is the ability to understand user intent, which is essential for directing customers effectively through the support funnel to the appropriate services. By employing sophisticated natural language processing (NLP) techniques, training LLM to perform RAG and machine learning models, businesses can precisely discern customer intents. This capability allows for the delivery of tailored, immediate responses. The paper further explores the methodologies employed, the advantages gained, and the challenges faced in the adoption of these advanced technologies in customer support systems.

The impact of emotional vs rational message framing on social media users' detection and sharing of misinformation: an experimental study

PurposeThis study aims to investigate the impact of message framing (emotional vs rational) on social media users' ability to accurately detect information and their intention to share messages about the COVID-19 vaccine.Design/methodology/approachUsing an experimental design approach, the authors recruited 600 adult participants via a crowdsourcing platform. Participants were randomly assigned to receive emotional or rational messages and their ability to accurately detect information and intention to share messages were assessed.FindingsThe results showed a significant multivariate effect of message framing on both the detection of accurate information and intention to share (p < 0.001). Participants who received emotional messages demonstrated better performance in the detection and sharing task than those who received rational messages. Gender and age also had significant main effects on the outcomes, with women performing better than men and younger participants performing better than older participants in detecting the accuracy of information. The interaction effects of the independent variables were not statistically significant (p = 0.098).Originality/valueThe findings highlight the importance of considering emotional factors in combating the spread of messages about the COVID-19 vaccine on social media. Practitioners responsible for social media content should strengthen the content review mechanism, with an emphasis on screening content with high emotional arousal.

Synchronizing real-time and high-precision LDoS defense of learning model-based in AIoT with programmable data plane, SDN

The availability of SD-AIoT is currently under complicated and serious cyber threats, especially Low-rate Denial-of-Service attacks. However, traditional defense schemes for such attacks with characteristics of high concealability and periodicity suffer from serious challenges with high detection difficulty, low accuracy of detection models, and inefficiency of mitigation approaches. In this paper, one novel cooperative defense scheme against hybrid LDoS attacks is proposed, which consists of a timely-response hardware-based Renyi Entropy edge checkpoint intent detection algorithm, the high-precision detection mechanism based on a hybrid deep learning model, and a Markov-chain-based differential rate-limiting mitigation strategy. The detection algorithm deployed at the edge checkpoint activates a hybrid CNN-RF-based deep learning model after filtering the intent information of the flows to detect which are malicious LDoS flows with high accuracy, where the multi-stage detection scheme not only extracts and learns the hidden features of the flow data, but also has better representation capabilities. Enhanced dynamic threshold-based whitelisting automatically adapts to the real-time state of the network environment to improve mitigation flexibility. Markov chain-based differential rate-limiting mitigation strategy reduces the packet loss error rate to mitigate network attacks promptly and ensures the continuation of network services. The results of several comparative experiments show that the proposed scheme detects LDoS attacks more accurately and mitigates them more effectively than traditional schemes.