User Behavior Model Research Articles

A unified user behavior model for trajectory-based tasks with different types of path constraints

A unified user behavior model for trajectory-based tasks with different types of path constraints

P2P Energy Trading through Prospect Theory, Differential Evolution, and Reinforcement Learning

Peer-to-peer (P2P) energy tradingis a decentralized energy market where local energyprosumersact as peers, trading energy among each other. Existing works in this area largely overlook the importance of user behavioral modeling and assume users’ sustained active participation and full compliance in the decision-making process. To overcome these unrealistic assumptions, and their deleterious consequences, in this article, we propose anautomatedP2P energy-trading framework that specifically considers the users’ perception by exploitingprospect theory. We formalize an optimization problem that maximizes the buyers’ perceived utility while matching energy production and demand. We prove that the problem is NP-hard and we propose a Differential Evolution-based Algorithm for Trading Energy (DEbATE) heuristic. Additionally, we propose two automated pricing solutions to improve the sellers’ profit based on reinforcement learning. The first solution, named Pricing mechanism with Q-learning and Risk-sensitivity (PQR), is based on Q-learning. Additionally, given the scalability issues ofPQR, we propose a Deep Q-Network-based algorithm calledProDQNthat exploits deep learning and a novel loss function rooted in prospect theory. Results based on real traces of energy consumption and production, as well as realistic prospect theory functions, show that our approaches achieve 26% higher perceived value for buyers and generate 7% more reward for sellers, compared to recent state-of-the-art approaches.

Facial expression-enhanced recommendation for virtual fitting rooms

With the development of Augmented Reality (AR) technology in the retail industry, virtual fitting room (VFR) are considered promising enhancement of e-commerce by providing users with an immersive environment to try on new products, especially fashion products. While allowing users having more vivid impression of products, virtual fitting rooms also offer sellers more channels to collect information on user preferences, which can be used to enhance recommender systems. This study proposes to leverage facial expression recognition technology together with fine-grained human-computer interactions in virtual fitting rooms to personalize product recommendations. This paper proposes a recommendation algorithm based on confidence setting, negative feedback sampling, and matrix factorization to model user behaviors in virtual fitting rooms. We conduct an experiment on 81 subjects to evaluate the proposed method. Experimental results show the proposed method outperforms existing methods using traditional behavior information. Our study provides a strong support to the value of AR in enhancing e-commerce.

Data adjusting strategy and optimized XGBoost algorithm for novel insider threat detection model

With the growth of access to internal system resources, the insider threat problem is emerging and can bring immeasurable losses to enterprises. In order to detect the hidden threats and guide the formulation of enterprise management strategy, it is important to analyze and understand the employee behavior. Thus, we suggest a user behavior analysis system framework, which mainly includes data processing, user behavior modeling and results analysis. Data Adjusting (DA) strategy and optimized eXtreme Gradient Boosting (XGBoost) model are utilized with the aim of full analysis small amount of feature information. The strategy for detecting suspicious behavior can be the following. Firstly, select initial suspicious data, misclassification data retention and combination sampling. Secondly, for further behavior model construction, an improved Particle Swarm Optimization algorithm based on ethnic randomized particles (ERPSO), which introduces Gaussian white noise with adjustable intensity into acceleration coefficients is given for searching the optimal XGBoost parameters. In addition, based on the designed DA strategy and the proposed ERPSO algorithm, we have also compared the results of the proposed methods with the current state-of-the-art methods. Experimental results show that the XGBoost optimized by the ERPSO (ERPSO-XGBoost) model has comprehensive performance, which proves the rationality and effectiveness of the insider behavior analysis system framework. Through a comprehensive understanding of insider behavior, the obvious characteristic behavior is found to adjust the management strategy and guide the behavior modeling, so as to prevent more losses in time.

Investigating the Role of SNS Design on Covid-19 Information Avoidance: The Perspectives of the Stimulus-Organism-Response Theory

During the COVID-19 pandemic, many organizations distributed important information through social networking sites (SNS). Unfortunately, the information can be hindered due to information avoidance, where users choose not to receive the information. While there is existing research on information avoidance, studies on how social media design can trigger information avoidance still need to be made available. The purpose of this study is to create a conceptual model of SNS user behavior based on the Stimulus-Organism-Response (SOR) theory, explaining the impact of digital interactivity (communication between humans and computers) on SNS users' decisions to avoid information. Therefore, in this study, an investigation was conducted on the role of SNS design in information avoidance, using S-O-R as the theoretical framework. A total of 332 respondents were included in the study. Using Covariance-Based Structural Equation Modeling (SEM-CB), we found that SNS design was related to information avoidance through the mediation of information overload and information anxiety. Based on these factors, it is believed that information avoidance arises as a negative response to the stimulus of using SNS, leading to user decision-making. Consequently, designers are recommended to create features that meet user needs and minimize negative impacts that could interfere with the user experience. The paper provides a more detailed explanation and discussion of the knowledge and practical implications.

Let Graph Be the Go Board: Gradient-Free Node Injection Attack for Graph Neural Networks via Reinforcement Learning

Graph Neural Networks (GNNs) have drawn significant attentions over the years and been broadly applied to essential applications requiring solid robustness or vigorous security standards, such as product recommendation and user behavior modeling. Under these scenarios, exploiting GNN's vulnerabilities and further downgrading its performance become extremely incentive for adversaries. Previous attackers mainly focus on structural perturbations or node injections to the existing graphs, guided by gradients from the surrogate models. Although they deliver promising results, several limitations still exist. For the structural perturbation attack, to launch a proposed attack, adversaries need to manipulate the existing graph topology, which is impractical in most circumstances. Whereas for the node injection attack, though being more practical, current approaches require training surrogate models to simulate a white-box setting, which results in significant performance downgrade when the surrogate architecture diverges from the actual victim model. To bridge these gaps, in this paper, we study the problem of black-box node injection attack, without training a potentially misleading surrogate model. Specifically, we model the node injection attack as a Markov decision process and propose Gradient-free Graph Advantage Actor Critic, namely G2A2C, a reinforcement learning framework in the fashion of advantage actor critic. By directly querying the victim model, G2A2C learns to inject highly malicious nodes with extremely limited attacking budgets, while maintaining a similar node feature distribution. Through our comprehensive experiments over eight acknowledged benchmark datasets with different characteristics, we demonstrate the superior performance of our proposed G2A2C over the existing state-of-the-art attackers. Source code is publicly available at: https://github.com/jumxglhf/G2A2C.

Decision-Making Context Interaction Network for Click-Through Rate Prediction

Click-through rate (CTR) prediction is crucial in recommendation and online advertising systems. Existing methods usually model user behaviors, while ignoring the informative context which influences the user to make a click decision, e.g., click pages and pre-ranking candidates that inform inferences about user interests, leading to suboptimal performance. In this paper, we propose a Decision-Making Context Interaction Network (DCIN), which deploys a carefully designed Context Interaction Unit (CIU) to learn decision-making contexts and thus benefits CTR prediction. In addition, the relationship between different decision-making context sources is explored by the proposed Adaptive Interest Aggregation Unit (AIAU) to improve CTR prediction further. In the experiments on public and industrial datasets, DCIN significantly outperforms the state-of-the-art methods. Notably, the model has obtained the improvement of CTR+2.9%/CPM+2.1%/GMV+1.5% for online A/B testing and served the main traffic of Meituan Waimai advertising system.

Towards a multi-modal Deep Learning Architecture for User Modeling

Deep learning has succeeded in various applications, including image classification and feature learning. However, there needs to be more research on its use in Intelligent Tutoring Systems or Serious Games, particularly in modeling user behavior during learning or gaming sessions using multi-modal data. Creating an effective user model is crucial for developing a highly adaptive system. To achieve this, it is necessary to consider all available data sources to inform the user’s current state. This study proposes a user-sensitive deep multi-modal architecture that leverages deep learning and user data to extract a rich latent representation of the user. The architecture combines a Long Short-Term Memory, a Convolutional Neural Network, and multiple Deep Neu-ral Networks to handle the multi-modality of data. The resulting model was evaluated on a public multi-modal dataset, achieving better results than state-of-the-art algorithms for a similar task: opinion polarity detection. These findings suggest that the latent representation learned from the data is useful in discriminating behaviors. This proposed solution can be applied in various contexts where user modeling using multi-modal data is critical for improving the user experience.

Research on Efficient Multi-Behavior Recommendation Method Fused with Graph Neural Network

Currently, most recommendation algorithms only use a single type of user behavior information to predict the target behavior. However, when browsing and selecting items, users generate other types of behavior information, which is important, but often not analyzed or modeled by traditional recommendation algorithms. This study aims to design a multi-behavior recommendation algorithm based on graph neural networks by analyzing multiple types of behavior information in users’ product purchasing process, to fully utilize multiple types of user behavior information. The algorithm models users, items, and user behavior in multiple dimensions by incorporating attention mechanisms and multi-behavior learning into graph neural networks, and solves the problem of imbalanced user behavior weights from the perspective of multi-task loss optimization. After experimental verification, we proposed that the multi-behavior graph attention network (MGAT) algorithm has better performance compared to four other classical recommendation algorithms on the Beibei and Taobao datasets. The results demonstrate that the multi-behavior recommendation algorithm based on graph neural networks has practicality in fully utilizing multiple types of user information, and can solve the problem of imbalanced user behavior weights to some extent.

Audience design and egocentrism in reference production during human-computer dialogue

Our current understanding of the mechanisms that underpin language production in human-computer dialogue (HCD) is sparse. What work there is in the field of human-computer interaction (HCI) supposes that people tend to adapt their language allocentrically, taking into account the perceived limitations of their partners, when talking to computers. Yet, debates in human-human dialogue (HHD) research suggest that people may also act egocentrically when producing language in dialogue. Our research aims to identify whether, similar to HHD, users also produce egocentric language within speech-based HCD interactions and how this behaviour compares to interaction with human dialogue partners. Such knowledge benefits the field of HCI by better understanding the mechanisms present in language production during HCD, which can be used to build more nuanced theories and models of user behaviour to inform research and design of speech interfaces. Through two controlled experiments using an adapted director-matcher task similar to those used in research on perspective-taking in psycholinguistics, we show that people do take the computer's perspective into account less (i.e. behave more egocentrically) during HCD than in HHD (Experiment 1). However, this egocentric effect is eliminated when computers are framed as separate interlocutors rather than computers integrated in the interactive system and where differences in perspective are made salient, leading to similar levels of perspective-taking as with human partners (Experiment 2). We discuss the findings, emphasising potential explanations for this effect, focusing on how egocentric and allocentric production processes may interact, along with the impact of partner roles and the division of labour in HCD as an underlying explanation for the effects seen.

A Reasonable Effectiveness of Features in Modeling Visual Perception of User Interfaces

Training data for user behavior models that predict subjective dimensions of visual perception are often too scarce for deep learning methods to be applicable. With the typical datasets in HCI limited to thousands or even hundreds of records, feature-based approaches are still widely used in visual analysis of graphical user interfaces (UIs). In our paper, we benchmarked the predictive accuracy of the two types of neural network (NN) models, and explored the effects of the number of features, and the dataset volume. To this end, we used two datasets that comprised over 4000 webpage screenshots, assessed by 233 subjects per the subjective dimensions of Complexity, Aesthetics and Orderliness. With the experimental data, we constructed and trained 1908 models. The feature-based NNs demonstrated 16.2%-better mean squared error (MSE) than the convolutional NNs (a modified GoogLeNet architecture); however, the CNNs’ accuracy improved with the larger dataset volume, whereas the ANNs’ did not: therefore, provided that the effect of more data on the models’ error improvement is linear, the CNNs should become superior at dataset sizes over 3000 UIs. Unexpectedly, adding more features to the NN models caused the MSE to somehow increase by 1.23%: although the difference was not significant, this confirmed the importance of careful feature engineering.

Understanding dynamics of polarization via multiagent social simulation

It is widely recognized that the Web contributes to user polarization, and such polarization affects not just politics but also peoples’ stances about public health, such as vaccination. Understanding polarization in social networks is challenging because it depends not only on user attitudes but also their interactions and exposure to information. We adopt Social Judgment Theory to operationalize attitude shift and model user behavior based on empirical evidence from past studies. We design a social simulation to analyze how content sharing affects user satisfaction and polarization in a social network. We investigate the influence of varying tolerance in users and selectively exposing users to congenial views. We find that (1) higher user tolerance slows down polarization and leads to lower user satisfaction; (2) higher selective exposure leads to higher polarization and lower user reach; and (3) both higher tolerance and higher selective exposure lead to a more homophilic social network.

Behavioral recommendation engine driven by only non-identifiable user data

Most recommendation systems utilize personal data to device personalized recommendations for users. Even though it seems favorable, security risks like data breaches are inevitable. This research proposes a novel reinforcement learning ‘approach’ to recommend users without collecting identifiable data. With only user activity on a session, our proposed method can model and track user behavior and formulate a recommendation system. We conclude that our algorithms demonstrate positive results in capturing user behavior without collecting private data of any kind from the user. The research is two folds. On one hand, we experiment using traditional reinforcement learning techniques (MDP, Q-learning), and on the other hand, we use deep reinforcement learning algorithms (DQN, DDQN, and D3QN) on a movie recommendation scenario. Interestingly, we observe that MDP and D3QN works comparatively better on movie recommendations.

Long Term Energy Savings Through User Behavior Modeling in Smart Homes

The Internet of Things (IoT) has enabled real-time monitoring of energy consumption in smart homes through sensors embedded in the surrounding environment. In the post-pandemic world, domestic energy management has gained importance due to increased work-from-home consumption, making data collection in a smart home a relevant IoT application with many potential energy savings. However, this information is difficult for most users to understand, and existing monitoring systems’ savings results degrade over time. To address these challenges, this study presents a novel approach for domestic energy consumption, production, and comfort perception using color-based dashboards enhanced for user feedback interaction. The approach includes the management of in-home appliances and comfort levels according to user preferences to attain long-term energy savings. The approach includes multiple appealing strategies such as 3D representation, mobile connectivity, utility integration, and dynamic information, to increase long-term engagement and provides quantitative data on energy savings achieved for one year, where the average energy consumption was reduced by 19%. It was found that the approach sustained user engagement over time, with users actively participating in energy conservation efforts. A community survey with 208 participants was also developed and studied where 69% of the enquired considered our approach more attractive than existing market solutions, and 79% considered it more useful than existing solutions. Regarding the real-time information presented on our approach, 81% of the participants strongly or totally agree that it can change users’ behaviors.

Deep Learning for User Behaviour Prediction Using Streaming Analytics

Streams of web user interactions reflect behaviour of customers or users of a web application through which a company is being operated online. The interactions may be in the form of visits to web components and even purchases made by users in case of e-Commerce applications. Modelling user behaviour can help the organizations to ascertain patterns of user behaviours and improve their products and services to meet their needs besides making promotional schemes. There are many existing methods for modelling user behaviour. However, of late, deep learning models are found to be more accurate and useful. In this paper a deep learning based framework is proposed for predicting web user behaviour from streams of user interactions. The framework is based on the mechanisms that exploit Recurrent Neural Network (RNN), one of the deep learning approaches, to learn from low-level features of sequential and streaming data. The mechanisms are used to model user interactions and predict the user behaviour with respect to purchasing items in future. In presence of plenty of items, item embeddings is explored for better results. In addition to this, attention mechanisms are employed to achieve RNN model interoperability. The empirical study revealed that the proposed framework is useful besides helping to evaluate different variants of attention mechanisms and item embeddings.

Research of marketing attribution and data visualization using google analytics

  The formation of user paths on the way to conversion, customer touch points with the brand, funnel and their evaluation are current issues in the field of marketing. Despite the rapid spread of technology, there are still no general established user guidelines for analyzing the marketing activities of enterprises in combination with the use of analysis tools. The main purpose of this study is the use of analysis tools in working with big data, as well as the ability to evaluate the implementation of tracking the user path from the moment the ad is displayed to the completion of targeted conversions in the final product. The article discusses one of the most advanced analytics systems - Google Analytics, also known as end-to-end analytics, which provides companies with opportunities to conduct basic analysis. Based on this tool, the article carried out work on the implementation of Google Analytics in the «Avtobys» application operating in the field of e-commerce. As a result of the work, unified recommendations for the implementation of an analytics system for applications were developed. Application performance criteria are defined. A model of user behavior in the application was modeled. Key words: Google Analytics, user behavior, marketing attribution, end-to-end analytics, funnel, user journey.

Recommending Profiles To Social Event Participation In Online Social Network Service Applications Using Deep Learning Techniques

Nowadays, organizing online Social events is growing especially in current pandemic situation due to corona. Thus social network services helps to organize the event oriented online social gathering. The events discussed on social networks can be associated with topics, locations, and time periods. Various literatures have employed clustering mechanism to group the profiles of the user of social networking applications. More research has been carried out in event based online network services as existing models leads to scalability and sparsity problems. Event-based online social networks are used to maintain interest-based groups with high relevancy rate, recommendation quality and predictive accuracy. In order to achieve the above goal, in this paper, we propose a novel framework named as Deep Influence Predict (DIP) which explores the features of Recurrent Neural Network in order identify the target or potential users through different patterns and behaviours of the profile on the social networking service applications. It learns multiple levels of representations and abstractions of the latent data through individual participation record. Further, it extracts the extrinsic and intrinsic properties of the profiles on their social connections and social effects. Specifically, it identifies the distinguishing social groups with different topics and categories as multifaceted interest in iterative process. Finally decision of recommendation for the event is integrated on outcomes of user behaviour model through personnel impact, social relation and equilibrium. Evaluation of the proposed model through various case studies has been implemented using hadoop architecture and validated across various measures such as accuracy on precision, Recall and f measure along scalability and Execution time.

Anticipated versus Actual Effects of Platform Design Change: A Case Study of Twitter's Character Limit

The design of online platforms is both critically important and challenging, as any changes may lead to unintended consequences, and it can be hard to predict how users will react. Here we conduct a case study of a particularly important real-world platform design change: Twitter's decision to double the character limit from 140 to 280 characters to soothe users' need to "cram" or "squeeze" their tweets, informed by modeling of historical user behavior. In our analysis, we contrast Twitter's anticipated pre-intervention predictions about user behavior with actual post-intervention user behavior: Did the platform design change lead to the intended user behavior shifts, or did a gap between anticipated and actual behavior emerge? Did different user groups react differently? We find that even though users do not "cram" as much under 280 characters as they used to under 140 characters, emergent "cramming" at the new limit seems to not have been taken into account when designing the platform change. Furthermore, investigating textual features, we find that, although post-intervention "crammed" tweets are longer, their syntactic and semantic characteristics remain similar and indicative of "squeezing". Applying the same approach as Twitter policy-makers, we create updated counterfactual estimates and find that the character limit would need to be increased further to reduce cramming that re-emerged at the new limit. We contribute to the rich literature studying online user behavior with an empirical study that reveals a dynamic interaction between platform design and user behavior, with immediate policy and practical implications for the design of socio-technical systems.

A Framework for Learning Event Sequences and Explaining Detected Anomalies in a Smart Home Environment

This paper presents a framework for learning event sequences for anomaly detection in a smart home environment. It addresses environment conditions, device grouping, system performance and explainability of anomalies. Our method models user behavior as sequences of events, triggered by interaction of the home residents with the Internet of Things (IoT) devices. Based on a given set of recorded event sequences, the system can learn the habitual behavior of the residents. An anomaly is described as deviation from that normal behavior, previously learned by the system. One key feature of our framework is the explainability of detected anomalies, which is implemented through a simple rule analysis.

A probabilistic perspective on nearest neighbor for implicit recommendation

Over the past years, the recommender systems community invented several novel approaches that reached better and better prediction accuracy. Sequential recommendation, such as music recommendation, has seen large improvements from neural network-based models such as recurrent neural networks or transformers. When no sequential information is available or not relevant, such as for book, movie, or product recommendation, however, the classic k-nearest neighbor algorithm appears to remain competitive, even when compared to much more sophisticated methods. In this paper, we attempt to explain the inner workings of the nearest neighbor using probabilistic tools, treating similarity as conditional probability and presenting a novel model for explaining and removing popularity bias. First, we provide a probabilistic formulation of similarity and the classic prediction formula. Second, by modeling user behavior as a combination of personal preference and global influence, we are able to explain the presence of popularity bias in the predictions. Finally, we utilize Bayesian inference to construct a theoretically grounded variant of the widely used inverse frequency scaling, which we use to mitigate the effect of popularity bias in the predictions. By replacing the formerly ad hoc choices of nearest neighbor with probabilistically founded counterparts, we are able to improve prediction accuracy over a variety of data sets and gain an increased understanding of the theory behind the method.