This paper proposes extended Long Short-Term Memory (LSTM) networks for the knowledge tracing task and employs explainable AI methods to address interpretability issues. Specifically, we developed an extended LSTM-based model to automatically diagnose students' knowledge states. We then leveraged three interpreting methods—gradient sensitivity, gradient*input, and Deep SHAP—to explain the model's predictions by computing input contributions. The results demonstrate that the proposed model outperforms DKT, and the three methods effectively explain its predictions. Additionally, we identified three key insights into the model's working mechanisms.

Artificial intelligence course has been required to take for compulsory education students in China. However, not all teachers and schools are fully prepared and ready. This is partially because of the lack of adequate teaching and learning resources, which requires a major expenditure of time and effort for schools and teachers to design and develop. To meet the challenge of lacking appropriate resources in teaching and learning AI from grade 1 to grade 9, we developed AI knowledge structure and instructional resources based on Chinese national curriculum for information science and technology. Our comprehensive AI syllabus contains 90 core concepts, 63 learning indicators, and 27 teaching and learning resources, which have been implemented. The resources have been taken as model courses in teacher training programs and an exemplary course has been implemented in primary schools that verified the effectiveness of our resources.

Deep learning-based knowledge tracing (DLKT) models have been regarded as the promising solution to estimate learners’ knowledge states and predict their future performance based on historical exercise records. However, the increasing complexity and diversity make DLKT models still difficult for users, typically including both learners and teachers, to understand models’ estimation results, directly hindering the model’s deployment and application. Previous studies have explored using methods from explainable artificial intelligence (xAI) to interpret DLKT models, but the methods have been limited in their generalizing capability and inefficient interpreting procedures. To address these limitations, we proposed a simple but efficient model-agnostic interpreting method, called Gradient*Input, to explain the predictions made by these models in two datasets. Comprehensive experiments have been conducted on the existing five DLKT models with representative neural network architectures. The experiment results showed that the method was effective in explaining the predictions of DLKT models. Further analysis of the interpreting results revealed that all five DLKT models share a similar rule in predicting learners’ item responses, and the role of skill and temporal information was found and discussed. We also suggested potential avenues for investigating the interpretability of DLKT models.

In response to the trend that artificial intelligence (AI) is becoming the main driver for social and economic development, enhancing the readiness of learners in AI is significant and important. The state council and the ministry of education of China put AI education for K-12 schools on a high priority in order to foster local AI talents and reduce educational disparities. However, the AI knowledge and technical skills are still limited for not only students but also the school teachers. Furthermore, many local schools in China, especially in the rural areas, are lack of the necessary software and hardware for teaching AI. Hence, we designed and implemented a structured series of AI courses, built on an online block-based visual programming platform. The AI courses are free and easily accessible for all. We have conducted the experimental classes in a local school and collected the results. The results show that the learners in general gained significant learning progress on AI knowledge comprehension, aroused strong interests in AI, and increased the degree of satisfaction towards the course. Especially, our practices significantly increased computational thinking of the students who were initially staying at a lower level.

Abstract How Deep Neural Networks (DNNs) best cope with the understanding of multimedia contents still remains an open problem, mainly due to two factors. First, conventional DNNs cannot effectively learn the representations of the images with sparse visual information. For example, the images describing knowledge concepts in textbooks. Second, existing DNNs cannot effectively capture the fine‐grained interactions between the images and text descriptions. To address these issues, we propose a deep Cross‐Media Grouping Fusion Network (CMGFN), which mainly has two distinctive properties: 1) CMGFN can effectively learn visual features from the images with sparse visual information. This is achieved by first progressively adjusting the attention of convolution filters to valuable visual regions, and then enhancing the use of key visual information in feature construction. 2) By a cross‐media grouping co‐attention mechanism, CMGFN can effectively use the interactions between visual features of different semantics and textual descriptions, to learn cross‐media features representing different fine‐grained semantics in different groups. Empirical studies demonstrate that CMGFN not only achieves state‐of‐the‐art performance on the multimedia documents containing sparse visual information, but also shows superior general applicability on other multimedia data, e.g., the multimedia fake news.

We design and implement an intelligent assistant, called PB-Advisor, to advise teachers and parents on students' problem behaviors. It utilizes a task-oriented dialogue system to identify the need deficiency underlying students' problem behaviors, and relies on a community question answering system to provide advice on typical problem behavior management. In addition, it also provides various learning resources, and illustrates the relations between influential factors on typical problem behaviors through data analysis. With PB-Advisor, teachers and parents without psychological expertise can easily find proper advice on students’ problem behaviors.

We propose and implement a novel intelligent tutoring system, called RadarMath, to support intelligent and personalized learning for math education. The system provides the services including automatic grading and personalized learning guidance. Specifically, two automatic grading models are designed to accomplish the tasks for scoring the text-answer and formula-answer questions respectively. An education-oriented knowledge graph with the individual learner’s knowledge state is used as the key tool for guiding the personalized learning process. The system demonstrates how the relevant AI techniques could be applied in today's intelligent tutoring systems.

The interest in artificial intelligence (AI) education is growing exponentially; nevertheless, how to learn about AI, particularly Natural Language Processing (NLP), has been a challenging problem for educators and researchers worldwide. This study used a graphical programming platform Snap! to facilitate learning by allowing learners to explore AI and its NLP techniques in class. Data from 18,452 logged events were collected and Lag Sequential Analysis (LSA) was used to examine how learners behaved and learned sequentially. Non-cognitive factors were used to group learners as detailed and subtle behavior sequences that did not occur by chance could be uncovered. The results showed that five groups of learners, that is Passive Learners, Performers, Adaptive Learners, Interested Learners, and Dedicated Learners. They presented varied learning behavior patterns, which should be considered further in designing personalized and intelligent learning platforms to support AI education.

We propose and implement a novel online learning system, called CogLearn, to support learner's self-awareness and reflective thinking, which urges a proper form of knowledge representation together with individual learner's cognitive status. We thus design and employ the machine learning techniques to estimate learner's cognitive status and identify educational relations to construct the desired knowledge representation, namely cognitive graph in our system. We further demonstrate the system by presenting two practical services, i.e., learning obstacle diagnosis and learning path planning, to demonstrate how the constructed cognitive graph effectively and adaptively supports individual system user's learning process.

We advocate for and presentTourSense, a framework for tourist identification and preference analytics using city-scale transport data (bus, subway, etc.). Our work is motivated by the observed limitations of utilizing traditional data sources (e.g., social media data and survey data) that commonly suffer from the limited coverage of tourist population and unpredictable information delay.TourSensedemonstrates how the transport data can overcome these limitations and provide better insights for different stakeholders, typically including tour agencies, transport operators, and tourists themselves. Specifically, we first propose a graph-based iterative propagation learning algorithm to recognize tourists from public commuters. Taking advantage of the trace data from the identified tourists, we then design a tourist preference analytics model to learn and predict their next tour, where an interactive user interface is implemented to ease the information access and gain the insights from the analytics results. Experiments with real-world datasets (from over 5.1 million commuters and their 462 million trips) show the promise and effectiveness of the proposed framework: the Macro and Micro F1 scores of the tourist identification system achieve 0.8549 and 0.7154, respectively, whereas the tourist preference analytics system improves the baselines by at least 23.53 and 11.44 percent in terms of precision and recall.

Knowledge tracing serves as the key technique in the computer supported education environment (e.g., intelligent tutoring systems) to model student's knowledge states. While the Bayesian knowledge tracing and deep knowledge tracing models have been developed, the sparseness of student's exercise data still limits knowledge tracing's performance and applications. In order to address this issue, we advocate for and propose to incorporate the knowledge structure information, especially the prerequisite relations between pedagogical concepts, into the knowledge tracing model. Specifically, by considering how students master pedagogical concepts and their prerequisites, we model prerequisite concept pairs as ordering pairs. With a proper mathematical formulation, this property can be utilized as constraints in designing knowledge tracing model. As a result, the obtained model can have a better performance on student concept mastery prediction. In order to evaluate this model, we test it on five different real world datasets, and the experimental results show that the proposed model achieves a significant performance improvement by comparing with three knowledge tracing models.

Motivated by the pressing need of educational applications with knowledge graph, we develop a system, called K12EduKG, to automatically construct knowledge graphs for K-12 educational subjects. Leveraging on heterogeneous domain-specific educational data, K12EduKG extracts educational concepts and identifies implicit relations with high educational significance. More specifically, it adopts named entity recognition (NER) techniques on educational data like curriculum standards to extract educational concepts, and employs data mining techniques to identify the cognitive prerequisite relations between educational concepts. In this paper, we present details of K12EduKG and demonstrate it with a knowledge graph constructed for the subject of mathematics.

We advocate for and introduce TRANSense, a framework for urban transportation service analytics that combines participatory smartphone sensing data with city-scale transportation-related transactional data (taxis, trains, etc.). Our work is driven by the observed limitations of using each data type in isolation: (a) commonly-used anonymous city-scale datasets (such as taxi bookings and GPS trajectories) provide insights into the aggregate behavior of transport infrastructure, but fail to reveal individual-specific transport experiences (e.g., wait times in taxi queues); while (b) mobile sensing data can capture individual-specific commuting-related activities, but suffers from accuracy and energy overhead challenges due to usage artefacts and lack of appropriate sensing triggers. TRANSense demonstrates how a judicious fusion of such disparate data sources can overcome these challenges and offer novel insights. We detail two examples: (a) Taxi Service Analyzer that provides accurate detection of commuter queuing for taxis and estimates their wait time, by using taxi trip records to identify potential taxi locations with high demand and subsequently selectively triggering mobile sensing-based queuing analytics on nearby commuters; and (b) Subway Boarding Analyzer that identifies instances when passengers fail to board arriving trains, by first estimating train arrivals from temporal patterns of passenger egress at station gantries, and then using mobile sensing-based analysis of commuter movement behavior on platforms. Experiments with real-world datasets (from over 20,000 taxis and 1.7 million commuters in Singapore) show the power of this approach: the taxi service analyzer detects commuter queuing with over 90 percent accuracy with negligible energy overhead and estimates wait times with error margins below 15 percent, whereas the subway boarding analyzer can detect failed boarding events with a precision of over 90 percent (more than thrice what is achievable through purely mobile sensing).

Public bus services are often planned in the context of urban planning. For a city with efficient and extensive network of public transportation system like Singapore, enhancing the existing coverage of bus service to meet the dynamic mobility needs of the population requires data mining approach. Specifically, frequent taxi rides between two locations at a period of time may suggest possible poor coverage of public transport service, if not lacking of the public transport service. In this paper, we describe a proof of concept effort to discover this weakness and its improvement in public transportation system via mining of taxi ride dataset. We cluster taxi rides dataset to determine some popular taxi rides in Singapore. From the clustered taxi rides, we filter and select only the clusters whose commuting via existing public transport are tortuous if not unreachable door-to-door. Based on the discovered travel pattern, we propose new bus routes that serve the passengers of these clusters. We formulate the bus planning problem as an optimization of directed cycle graph, and present it's preliminary solution and results. We showcase our idea in the case of Singapore.

Clustering of a large amount of taxi GPS mobility data helps to understand the spatio-temporal dynamics for the applications of urban planning and transportation. In this paper we cluster the origin-destination pairs of the passenger taxi rides to provide useful insight into the city mobility patterns, urban hot-spots, road network usage and general patterns of the crowd movement within the city of Singapore. We perform experiments on a large scale Singapore taxi dataset consisting of more than 10 million passenger origin-destination GPS points. We use the clusi VAT sampling scheme to obtain the sample trips which return coarse clusters describing the major crowd movement and reduce the data points that are not captured by the coarse clusters and may bring in noises during fine-grained clustering. After the sampling step we use the well known density based clustering algorithm DBSCAN to find cluster structure in the sampled data points and later extend it to the rest of the dataset using nearest prototype rule. We report 24 trip clusters from the dataset which are compact enough to draw meaningful conclusions about the city mobility patterns and the number of trips in each cluster is large enough to be representative of the general traffic movement.

We contend that context information of Internet clients can help to efficiently manage a variety of underlying resources for different Internet services and systems. We therefore propose a resource distribution framework that provides quality of experience (QoE) aware service differentiation, which means that starving clients are prioritized in resource allocation to enhance the corresponding end-user's QoE. The framework also actively motivates each Internet client to consistently provide its actual context information and to adopt moderate competition policies, given that all clients are selfish but rational in nature. We analyze the Internet client's behavior by formulating a non-cooperative game and prove that the framework guides all clients (game players) towards a unique Nash equilibrium. Furthermore, we prove that the distribution results computed by the framework maximize a social welfare function. Throughout this paper, we demonstrate the motivation, operation and performance of the framework by presenting a Web system example, which leverages on the advanced context information deduced by a context-aware system.

Taxi waiting queues or passenger waiting queues usually reflect the imbalance between taxi supply and demand, which consequently decrease a city’s trac system productivity and commuters’ satisfaction. In this paper, we present a queue detection and analysis system to conduct analytics on both taxi and passenger queues. The system utilizes the event-driven taxi traces and the taxi state transition knowledge to detect queue locations at a coordinate level and subsequently identify 4 di↵erent types of queue context (e.g., only passengers queuing or only taxis queuing). More specifically, it adopts the novel and easy-to-implement algorithms to selectively extract taxi pickup events and their critical features. The extracted taxi pickup locations are then used to detect queue locations, and the extracted critical features are used to infer queue context. The extensive empirical evaluations, which run on daily 12.4 million taxi trace records from nearly 15000 taxis in Singapore, demonstrate the high accuracy and stability of the queue analytics results. Finally, we discuss the real world deployment issues and the gained insights from the queue analysis results.

We present QueueVadis, a system that addresses the problem of estimating, in real-time, the properties of queues at commonplace urban locations, such as coffee shops, taxi stands and movie theaters. Abjuring the use of any queuing-specific infrastructure sensors, QueueVadis uses participatory mobile sensing to detect both (i) the individual-level queuing episodes for any arbitrarily-shaped queue (by a characteristic locomotive signature of short bursts of "shuffling forward" between periods of "standing") and (ii) the aggregate-level queue properties (such as expected wait or service times) via appropriate statistical aggregation of multi-person data. Moreover, for venues where multiple queues are too close to be separated via location estimates, QueueVadis also uses a novel disambiguation technique to separate users into multiple distinct queues. User studies, performed with 138 cumulative total users observed at 23 different real-world queues across Singapore and Japan, show that QueueVadis is able to (a) identify all individual queuing episodes, (b) predict service and wait times fairly accurately (with median estimation errors in the 10%--20% range), independent of the queue's shape, (c) separate users in multiple proximate queues with close to 80% accuracy and (d) provide reasonable estimates when the participation rate (the fraction of QueueVadis-equipped people in the queue) is modest.

Recently there has been increasing interest in building networks with Ambient Intelligence (AmI), which incorporates the user-centricity and context awareness. However, both the Internet TCP/IP protocol stack and the seven-layer OSI reference model are not suitable for AmI networks, because they do not specifically take the end-user requirements into consideration in their architecture design. Under the client-server architecture, we propose to explicitly take the end-user into account by defining a new layer called User Layer above the traditional application layer. The User Layer empowers the end-users to influence network performance based on their interaction activities with the networks. We adopt the Model Human Processor (MHP) approach for building the User Model. After that we present an exemplary User Layer implementation to illustrate how the User Layer interacts with the underlying protocol stack and improves end-user's satisfaction with network performance.

This paper addresses building networks that emphasizes on user-centric human computer interaction and context awareness. To achieve this user-centric intelligent network goal, we propose to explicitly take the end-user into account by defining a new layer called the User Layer above the traditional application layer. By exposing some lower layer information to the end-user, the new User Layer establishes a feedback loop between the end-user and the underlying network infrastructure, empowering the end-user to control and influence network performance based on his own behavior and preferences. A cross-layer design approach using a shared database between the different lower layers is adopted. To illustrate the User Layer in action, we present an exemplary implementation of the User Layer, which can dynamically allocate network resources by leveraging on the TCP flow control mechanism. We evaluated network performance via simulation and show that such a design improves the user perceived quality of service (QoS).