User Intent Detection Research Articles

Divide–Aggregate Heterogeneous Hypergraph for large-scale user intention detection

Detecting the intention of users in the information retrieval system is a critical task affecting the user experience. Confronting the challenge of retrieval datasets can be extremely large, complex, and heterogeneous correlated, we propose a framework named DAHH, which utilizes the Divide and Aggregate strategy to incorporate Heterogeneous Hypergraph for large-scale user retrieval intention detection. In the proposed framework, the large-scale retrieval dataset is first divided into multiple subsets with a solvable scale of samples for the model. Then consistent rules are used in all the subsets to construct heterogeneous hypergraphs, and the hypergraph convolution operation is conducted independently on each hypergraph to generate local high-order topologic enhanced vertex representations. Aggregation is applied to correlate all the local hypergraphs at different levels to generate global high-order heterogeneous enhanced representations. In the end, local and global high-order representations are combined to predict user intentions. The proposed framework is capable of dealing with heterogeneous and high-order correlations among the large-scale user retrieval data. We conduct extensive experiments on our collected real-world search engine dataset with millions of samples as well as several widely used public datasets to demonstrate the across-the-board superior performance of DAHH to other state-of-the-art models.

Continuous A-Mode Ultrasound-Based Prediction of Transfemoral Amputee Prosthesis Kinematics Across Different Ambulation Tasks.

Volitional control systems for powered prostheses require the detection of user intent to operate in real life scenarios. Ambulation mode classification has been proposed to address this issue. However, these approaches introduce discrete labels to the otherwise continuous task that is ambulation. An alternative approach is to provide users with direct, voluntary control of the powered prosthesis motion. Surface electromyography (EMG) sensors have been proposed for this task, but poor signal-to-noise ratios and crosstalk from neighboring muscles limit performance. B-mode ultrasound can address some of these issues at the cost of reduced clinical viability due to the substantial increase in size, weight, and cost. Thus, there is an unmet need for a lightweight, portable neural system that can effectively detect the movement intention of individuals with lower-limb amputation. In this study, we show that a small and lightweight A-mode ultrasound system can continuously predict prosthesis joint kinematics in seven individuals with transfemoral amputation across different ambulation tasks. Features from the A-mode ultrasound signals were mapped to the user's prosthesis kinematics via an artificial neural network. Predictions on testing ambulation circuit trials resulted in a mean normalized RMSE across different ambulation modes of 8.7 ± 3.1%, 4.6 ± 2.5%, 7.2 ± 1.8%, and 4.6 ± 2.4% for knee position, knee velocity, ankle position, and ankle velocity, respectively. This study lays the foundation for future applications of A-mode ultrasound for volitional control of powered prostheses during a variety of daily ambulation tasks.

A Novel Hybrid Approach for Intent Creation and Detection Using K-Means-Based Topic Clustering and Heuristic-Based Capsule Network

Background: The social media revolution has offered new facilities and opportunities to the online community to communicate their intentions, opinions, and views regarding products, services, policies, and events. The identification of intent focuses on the detection of intents from user reviews, that is, whether the specific review of the user includes intention or not. Intent mining is also named intent identification which helps business organizations to identify the purchase intentions of users. However, detecting user intentions encoded in text queries is a complicated task in several Natural Language Processing (NLP) applications such as robots, smart agents, personal assistants, and search engines. The existing research works have discovered the utilization of several machine learning techniques to detect the intents from queries of users. Most works consider intent detection as a classification problem, with utterances as predefined intents. Research question: Whether the researcher resolves the detection of user intentions encoded in text queries? How the researcher solves the existing challenges based on intent mining? Purpose: The main contribution of the research is to design and implement intent detection using topic clustering and deep learning. Methodology: Initially, the dataset related to diverse queries is gathered. Then, the label creation is performed by clustering. The clustering is performed by a k-means clustering model with a cosine similarity function. Once the clustering is performed for different queries, the label is created, which is used to train the network under the detection process. For the detection, this paper uses a Heuristic-based Capsule Network (H-CapNet) that could perform the intention for a new query. The hybrid meta-heuristic algorithm with Escaping Energy searched Grey–Harris Hawks Algorithm (EEG-HHA) is used for improving the capsule network. Validation: Experimental analysis shows that the developed method has superior performance in evaluating standard datasets with other approaches. Results: From the simulation results, the accuracy of the developed EEG-HHA-CapNet for dataset 1 is secured at 3%, 1.6%, 2%, and 1.1% increased than PSO-CapNet, WOA-CapNet, HHO-CapNet, and GWO-CapNet. Conclusion: Thus, the designed user intent detection models reveal their more advanced performance based on the diverse performance and error metrics for datasets 1 and 2.

Improving User Intent Detection in Urdu Web Queries with Capsule Net Architectures

Detecting the communicative intent behind user queries is critically required by search engines to understand a user’s search goal and retrieve the desired results. Due to increased web searching in local languages, there is an emerging need to support the language understanding for languages other than English. This article presents a distinctive, capsule neural network architecture for intent detection from search queries in Urdu, a widely spoken South Asian language. The proposed two-tiered capsule network utilizes LSTM cells and an iterative routing mechanism between the capsules to effectively discriminate diversely expressed search intents. Since no Urdu queries dataset is available, a benchmark intent-annotated dataset of 11,751 queries was developed, incorporating 11 query domains and annotated with Broder’s intent taxonomy (i.e., navigational, transactional and informational intents). Through rigorous experimentation, the proposed model attained the state of the art accuracy of 91.12%, significantly improving upon several alternate classification techniques and strong baselines. An error analysis revealed systematic error patterns owing to a class imbalance and large lexical variability in Urdu web queries.

Development of Dialogue Management System for Banking Services

Rapid increase in conversational AI and user chat data lead to intensive development of dialogue management systems (DMS) for various industries. Yet, for low-resource languages, such as Azerbaijani, very little research has been conducted. The main purpose of this work is to experiment with various DMS pipeline set-ups to decide on the most appropriate natural language understanding and dialogue manager settings. In our project, we designed and evaluated different DMS pipelines with respect to the conversational text data obtained from one of the leading retail banks in Azerbaijan. In the work, the main two components of DMS—Natural language Understanding (NLU) and Dialogue Manager—have been investigated. In the first step of NLU, we utilized a language identification (LI) component for language detection. We investigated both built-in LI methods such as fastText and custom machine learning (ML) models trained on the domain-based dataset. The second step of the work was a comparison of the classic ML classifiers (logistic regression, neural networks, and SVM) and Dual Intent and Entity Transformer (DIET) architecture for user intention detection. In these experiments we used different combinations of feature extractors such as CountVectorizer, Term Frequency-Inverse Document Frequency (TF-IDF) Vectorizer, and word embeddings for both word and character n-gram based tokens. To extract important information from the text messages, Named Entity Extraction (NER) component was added to the pipeline. The best NER model was chosen among conditional random fields (CRF) tagger, deep neural networks (DNN), models and build in entity extraction component inside DIET architecture. Obtained entity tags fed to the Dialogue Management module as features. All NLU set-ups were followed by the Dialogue Management module that contains a Rule-based Policy to handle FAQs and chitchats as well as a Transformer Embedding Dialogue (TED) Policy to handle more complex and unexpected dialogue inputs. As a result, we suggest a DMS pipeline for a financial assistant, which is capable of identifying intentions, named entities, and a language of text followed by policies that allow generating a proper response (based on the designed dialogues) and suggesting the best next action.

Generation of Gait Events with a FSR Based Cane Handle.

Gait analysis has many applications, and specifically can improve the control of prosthesis, exoskeletons, or Functional Electrical Stimulation systems. The use of canes is common to complement the assistance in these cases, and the synergy between upper and lower limbs can be exploited to obtain information about the gait. This is interesting especially in the case of unilateral assistance, for instance in the case of one side lower limb exoskeletons. If the cane is instrumented, it can hold sensors that otherwise should be attached to the body of the impaired user. This can ease the use of the assistive system in daily life as well as its acceptance. Moreover, Force Sensing Resistors (FSRs) are common in gait phase detection systems, and force sensors are also common in user intention detection. Therefore, a cane that incorporates FSRs on the handle can take advantage from the direct interface with the human and provide valuable information to implement real-time control. This is done in this paper, and the results confirm that many events are detected from variables derived from the readings of the FSRs that provide rich information about gait. However, a large inter-subject variability points to the need of tailored control systems.

Few-shot Learning for Multi-label Intent Detection

In this paper, we study the few-shot multi-label classification for user intent detection. For multi-label intent detection, state-of-the-art work estimates label-instance relevance scores and uses a threshold to select multiple associated intent labels. To determine appropriate thresholds with only a few examples, we first learn universal thresholding experience on data-rich domains, and then adapt the thresholds to certain few-shot domains with a calibration based on nonparametric learning. For better calculation of label-instance relevance score, we introduce label name embedding as anchor points in representation space, which refines representations of different classes to be well-separated from each other. Experiments on two datasets show that the proposed model significantly outperforms strong baselines in both one-shot and five-shot settings.

Relationship Between Muscular Activity and Assistance Magnitude for a Myoelectric Model Based Controlled Exosuit.

The growing field of soft wearable exosuits, is gradually gaining terrain and proposing new complementary solutions in assistive technology, with several advantages in terms of portability, kinematic transparency, ergonomics, and metabolic efficiency. Those are palatable benefits that can be exploited in several applications, ranging from strength and resistance augmentation in industrial scenarios, to assistance or rehabilitation for people with motor impairments. To be effective, however, an exosuit needs to synergistically work with the human and matching specific requirements in terms of both movements kinematics and dynamics: an accurate and timely intention-detection strategy is the paramount aspect which assume a fundamental importance for acceptance and usability of such technology. We previously proposed to tackle this challenge by means of a model-based myoelectric controller, treating the exosuit as an external muscular layer in parallel to the human biomechanics and as such, controlled by the same efferent motor commands of biological muscles. However, previous studies that used classical control methods, demonstrated that the level of device's intervention and effectiveness of task completion are not linearly related: therefore, using a newly implemented EMG-driven controller, we isolated and characterized the relationship between assistance magnitude and muscular benefits, with the goal to find a range of assistance which could make the controller versatile for both dynamic and static tasks. Ten healthy participants performed the experiment resembling functional daily activities living in separate assistance conditions: without the device's active support and with different levels of intervention by the exosuit. Higher assistance levels resulted in larger reductions in the activity of the muscles augmented by the suit actuation and a good performance in motion accuracy, despite involving a decrease of the movement velocities, with respect to the no assistance condition. Moreover, increasing torque magnitude by the exosuit resulted in a significant reduction in the biological torque at the elbow joint and in a progressive effective delay in the onset of muscular fatigue. Thus, contrarily to classical force and proportional myoelectric schemes, the implementation of an opportunely tailored EMG-driven model based controller affords to naturally match user's intention detection and provide an assistance level working symbiotically with the human biomechanics.

Classification of the User's Intent Detection in Ecommerce systems – Survey and Recommendations

Classification of the User's Intent Detection in Ecommerce systems – Survey and Recommendations

사용자 의도감지를 위한 소방대원 근력 지원 장치 조작기 개발

This paper describes the development of a power assistive device controller with user intention detection for fire fighters. In order to detect the intention of users, an F/T sensor frame was designed for the power assistive device controller. Using the numerical approach, each directional strain value of the F/T sensor frame was evaluated singly to determine the optimum point to mount the strain gauge under varying load conditions. The numerical analysis was conducted using the commercial program Ansys v11.0. The finite element model for the F/T sensor frame consisted of 37,547 elements and 157,154 nodes. A sensor bonding device and calibration jig were designed for the F/T sensor frame. In an effort to obtain the decoupling matrix for the F/T sensor frame of the proposed power assist device, calibration tests were conducted in the x-direction, y-direction, z-direction, My-direction and Mz-direction. In addition, the operating system was tested using the power assistive device controller that comprised of the F/T sensor frame.

Towards the Development of a Learning-Based Intention Classification Framework for Pushrim-Activated Power-Assisted Wheelchairs.

There has been a growth in the design and use of power assist devices for manual wheelchairs (MWCs) to alleviate the physical load of MWC use. A pushrim-activated power-assisted wheel (PAPAW) is an example of a power assist device that replaces the conventional wheel of a MWC. Although the use of PAPAWs provides some benefits to MWC users, it can also cause difficulties in maneuvering the wheelchair. In this research, we examined the characteristics of wheelchair propulsion when using manual and powered wheels. We used the left and right wheels' angular velocity to calculate the linear and angular velocity of the wheelchair. Results of this analysis revealed that the powered wheel's controller is not optimally designed to reflect the intentions of a wheelchair user. To address some of the challenges with coordinating the pushes on PAPAWs, we proposed the design of a user-intention detection framework. We used the kinematic data of MWC experiments and tested six supervised learning algorithms to classify one of four movements: "not moving", "moving straight forward", "turning left", and "turning right". We found that all the classification algorithms determined the type of movement with high accuracy and low computation time. The proposed intention detection framework can be used in the design of learning-based controllers for PAPAWs that take into account the individualized characteristics of wheelchair users. Such a system may improve the experience of PAPAW users.

Review of Intent Diversity in Information Retrieval : Approaches, Models and Trends

The fast increasing volume of information databases made some difficulties for a user to find the information that they need. Its important for researchers to find the best method for challenging this problem. user intention detection can be used to increase the relevancies of information delivered from the information retrieval system. This research used a systematic mapping process to identify what area, approaches, and models that mostly used to detect user intention in information retrieval in four years later. the result of this research identified that item-based approach is still the most approach researched by researchers to identify intent diversity in information retrieval. The used of item-based approach still increasing from 2015 until 2017. 34% paper used topic models in their research. It means that Topic models still the necessary models explored by the researchers in this study.

Strokes of insight: User intent detection and kinematic compression of mouse cursor trails

Web users often have a specific goal in mind comprising various stages that are reflected, as executed, by their mouse cursor movements. Therefore, is it possible to detect automatically which parts of those movements bear any intent and discard the parts that have no intent? Can we estimate the intent degree of the non-discarded parts? To achieve this goal, we tap into the Kinematic Theory and its associated Sigma-Lognormal model (ΣΛM). According to this theory, the production of a mouse cursor movement requires beforehand the instantiation of an action plan. The ΣΛM models such an action plan as a sequence of strokes’ velocity profiles, one stroke at a time, providing thus a reconstruction of the original mouse cursor movement. When a user intent is clear, the pointing movement is faster and the cursor movement is reconstructed almost perfectly, while the reverse is observed when the user intent is unclear.We analyzed more than 10,000 browsing sessions comprising about 5 million of data points, and compared different segmentation techniques to detect discrete cursor chunks that were then reconstructed with the ΣΛM. Our main contribution is thus a novel methodology to automatically tell chunks with and without intention apart. We also contribute with kinematic compression, a novel application to compress mouse cursor data while preserving most of the original information. Ultimately, this work enables a deeper understanding of mouse cursor movements production, providing an informed means to gain additional insight about users’ browsing behavior.

A Conversational Agent to Encourage Willingness to Communicate in the Context of English as a Foreign Language

Abstract We propose an embodied conversational agent based on the willingness to communicate (WTC) model in L2 to help increase WTC in the context of English as a Foreign Language (EFL) by providing users with various daily conversation contexts. To simulate realistic and efficient conversations, we adopted a semantic approach in the response generation and created a system with flexible and adaptable domain knowledge, user's intent detection, and mixed-initiative conversation strategy. Our evaluation of the proposed system demonstrated its potential to increase WTC in the EFL context.

Walking intent detection algorithm for paraplegic patients using a robotic exoskeleton walking assistant with crutches

This paper deals with a walking intent detection algorithm for paraplegic patients using a robotic exoskeleton walking assistant with crutches. User intent detection is one of the most important factors for a robotic exoskeleton to recognize a user’s intent regarding robot operation and in order to conduct proper operation. For a robotic exoskeleton walking assistant, walking intent detection means that the robot recognizes the best time to start walking behaviors and which behavior should be executed from the user’s intent. To enable this detection, an algorithm derived from a finite Moore Automaton for Robot Behavior (MARB) has been developed. Also, a walking stability criterion model is presented for use in the proposed algorithm as a trigger signal for walking. The proposed algorithm and the walking stability criterion modeling were verified using the robotic exoskeleton walking assistant, ROBIN.