A Theory-Driven Deep Learning Method for Voice Chat–Based Customer Response Prediction

2023 | Information Systems Research | Citations: 0

Authors: Chen, Gang; Xiao, Shuaiyong; Zhang, Chenghong; Zhao, Huimin

Abstract: As artificial intelligence and digitalization technologies are flourishing real- ... Expand

Abstract: As artificial intelligence and digitalization technologies are flourishing real-time, online interaction–based commercial modes, exploiting customers’ purchase intention implied in online interaction processes may foster huge business opportunities. In this study, we target the task of voice chat–based customer response prediction in an emerging online interaction–based commercial mode, the invite-online-and-experience-in-store mode. Prior research shows that satisfaction, which can be revealed by the discrepancy between prior expectation and actual experience, is a key factor to disentangle customers’ purchase intention, whereas black-box deep learning methods empirically promise us advantageous capabilities in dealing with complex voice data, for example, text and audio information incorporated in voice chat. To this end, we propose a theory-driven deep learning method that enables us to (1) learn customers’ personalized product preferences and dynamic satisfaction in the absence of their profile information, (2) model customers’ actual experiences based on multiview voice chat information in an interlaced way, and (3) enhance the customer response prediction performance of a black-box deep learning model with theory-driven dynamic satisfaction. Empirical evaluation results demonstrate the advantageous prediction performance of our proposed method over state-of-the-art deep learning alternatives. Investigation of cumulative satisfaction reveals the collaborative predictive roles of theory-driven dynamic satisfaction and deep representation features for customer response prediction. Explanatory analysis further renders insights into customers’ personalized preferences and dynamic satisfaction for key product attributes. History: Yong Tan, Senior Editor; Jingjing Zhang, Associate Editor. Funding: This work was supported in part by the National Natural Science Foundation of China [Grants 71971067 and 72271059] and the China Postdoctoral Science Foundation [Grant 2022M722394]. Supplemental Material: The online appendix is available at https://doi.org/10.1287/isre.2022.1196. Collapse

Topics: customer satisfaction willingness to pay decision making marketing information system customer service

Methods: deep learning experiment machine learning BERT theory development

Automated Deductive Content Analysis of Text: A Deep Contrastive and Active Learning Based Approach

2023 | International Conference on Information Systems | Citations: 0

Authors: De Oliveira Silveira, Alysson; Ray, Arindam; Ebrahimi, Mohammadreza; Bhattacherjee, Anol

Abstract: Content analysis traditionally involves human coders manually combing through t ... Expand

Abstract: Content analysis traditionally involves human coders manually combing through text documents to search for relevant concepts and categories. However, this approach is time-intensive and not scalable, particularly for secondary data like social media content, news articles, or corporate reports. To address this problem, the paper presents an automated framework called Automated Deductive Content Analysis of Text (ADCAT) that uses deep learning-based semantic techniques, an ontology of validated construct measures, a large language model, human-in-the-loop disambiguation, and a novel augmentation-based weighted contrastive learning approach for improved language representations, to build a scalable approach for deductive content analysis. We demonstrate the effectiveness of the proposed approach to identify firm innovation strategies from their 10-K reports to obtain inferences reasonably close to human coding. Collapse

Topics: knowledge representation human in the loop human computer interaction innovation management organizational innovation

Methods: ontological modelling machine learning language model large language model qualitative content analysis

Incorporating Multiple Knowledge Sources for Targeted Aspect-based Financial Sentiment Analysis

2023 | ACM Transactions on Management Information Systems | Citations: 0

Authors: Du, Kelvin; Xing, Frank; Cambria, Erik

Abstract: Combining symbolic and subsymbolic methods has become a promising strategy as re ... Expand

Abstract: Combining symbolic and subsymbolic methods has become a promising strategy as research tasks in AI grow increasingly complicated and require higher levels of understanding. Targeted Aspect-based Financial Sentiment Analysis (TABFSA) is an example of such complicated tasks, as it involves processes like information extraction, information specification, and domain adaptation. However, little is known about the design principles of such hybrid models leveraging external lexical knowledge. To fill this gap, we define anterior, parallel, and posterior knowledge integration and propose incorporating multiple lexical knowledge sources strategically into the fine-tuning process of pre-trained transformer models for TABFSA. Experiments on the Financial Opinion mining and Question Answering challenge (FiQA) Task 1 and SemEval 2017 Task 5 datasets show that the knowledge-enabled models systematically improve upon their plain deep learning counterparts, and some outperform state-of-the-art results reported in terms of aspect sentiment analysis error. We discover that parallel knowledge integration is the most effective and domain-specific lexical knowledge is more important according to our ablation analysis. Collapse

Topics: microblog domain-specific language Twitter social media knowledge graph

Methods: sentiment analysis RoBERTa transformer model BERT XLNet

Live streaming recommendations based on dynamic representation learning

2023 | Decision Support Systems | Citations: 0

Authors: Gao, Ge; Liu, Hongyan; Zhao, Kang

Abstract: As an emerging form of social media, live streaming services (e.g., Twitch and C ... Expand

Abstract: As an emerging form of social media, live streaming services (e.g., Twitch and Clubhouse) allow users to interact with hosts and peers in real time while enjoying shows or participating in discussions. These platforms are also dynamic, with shows or discussions changing quickly inside a room and users frequently switching between rooms. To improve user engagement and experience on such platforms, we design a new recommendation model named Dynamic Representations for Live Streaming Rooms (DRIVER) to provide room recommendations. Guided by the Integrated Framework for Consumer Path Modeling and the social affordance theory, DRIVER infers dynamic representations of live streaming rooms by leveraging users’ behavior paths in entering, staying in, and leaving rooms. One contribution of our model is a new and efficient dynamic learning framework to model instantaneous and ever-changing inter-room relationships by considering individual users’ behavior paths after leaving a room. Also supported by social affordance theory, another methodological novelty of our model is to capture dynamic characteristics of a room by incorporating features of the current audience inside the room. Experiments on real-world datasets from two different types of live streaming platforms demonstrate that DRIVER outperforms state-of-the-art representation learning methods and sequential recommender systems. The proposed method also has implications for recommender system design in other contexts, in which items are characterized by users’ dynamic behavior paths and ongoing social interactions. Collapse

Topics: live streaming platform recommender system user experience knowledge sharing systems design

Methods: experiment sensitivity analysis ablation analysis descriptive statistic deep learning

Theories: theory of affordance

Graph Learning of Multifaceted Motivations for Online Engagement Prediction in Counter-party Social Networks

2023 | International Conference on Information Systems | Citations: 0

Authors: Hu, Manting; LIN, Qingyuan; Zhang, Denghui; Lu, Angela; Junming, Liu; Xiong, Hui

Abstract: Social media has emerged as an essential venue to invigorate online political e ... Expand

Abstract: Social media has emerged as an essential venue to invigorate online political engagement. However, political engagement is multifaceted and impacted by both individuals' selfmotivation and social influence from peers and remains challenging to model in a counterparty network. Therefore, we propose a counter-party graph representation learning model to study individuals' intrinsic and extrinsic motivations for online political engagement. Firstly, we capture users' intrinsic political interests providing self-motivation from a user-topic network. Then, we encode how users cast influence on others from the inner-/counter-party through a user-user network. With the learned embedding of intrinsic and extrinsic motivations, we model the interactions between these two facets and utilize the dependency by deep sequential model decoding. Finally, extensive experiments using Twitter data related to the 2020 U.S. presidential election and the 2019 HK protests validate the model's predictive power. This study has implications for online political engagement, political participation, and political polarization. Collapse

Topics: social network Twitter social media user-generated content intrinsic motivation

Methods: machine learning experiment machine learning method computational algorithm ablation analysis

Theories: motivation theory

LiMS-Net: A Lightweight Multi-Scale CNN for COVID-19 Detection from Chest CT Scans

2023 | ACM Transactions on Management Information Systems | Citations: 0

Authors: Joshi, Amogh Manoj; Nayak, Deepak Ranjan; Das, Dibyasundar; Zhang, Yudong

Abstract: Recent years have witnessed a rise in employing deep learning methods, especiall ... Expand

Abstract: Recent years have witnessed a rise in employing deep learning methods, especially convolutional neural networks (CNNs) for detection of COVID-19 cases using chest CT scans. Most of the state-of-the-art models demand a huge amount of parameters which often suffer from overfitting in the presence of limited training samples such as chest CT data and thereby, reducing the detection performance. To handle these issues, in this paper, a lightweight multi-scale CNN called LiMS-Net is proposed. The LiMS-Net contains two feature learning blocks where, in each block, filters of different sizes are applied in parallel to derive multi-scale features from the suspicious regions and an additional filter is subsequently employed to capture discriminant features. The model has only 2.53M parameters and therefore, requires low computational cost and memory space when compared to pretrained CNN architectures. Comprehensive experiments are carried out using a publicly available COVID-19 CT dataset and the results demonstrate that the proposed model achieves higher performance than many pretrained CNN models and state-of-the-art methods even in the presence of limited CT data. Our model achieves an accuracy of 92.11% and an F1-score of 92.59% for detection of COVID-19 from CT scans. Further, the results on a relatively larger CT dataset indicate the effectiveness of the proposed model. Collapse

Topics: covid 19

Methods: computed tomography ImageNet ablation analysis experiment machine learning

ROLEX: A Novel Method for Interpretable Machine Learning Using Robust Local Explanations

2023 | Management Information Systems Quarterly | Citations: 2

Authors: Kim, Buomsoo (Raymond); Srinivasan, Karthik; Kong, Sung Hye; Kim, Jung Hee; Shin, Chan Soo; Ram, Sudha

Abstract: Recent developments in big data technologies are revolutionizing the field of h ... Expand

Abstract: Recent developments in big data technologies are revolutionizing the field of healthcare predictive analytics (HPA), enabling researchers to explore challenging problems using complex prediction models. Nevertheless, healthcare practitioners are reluctant to adopt those models as they are less transparent and accountable due to their black-box structure. We believe that instance-level, or local, explanations enhance patient safety and foster trust by enabling patient-level interpretations and medical knowledge discovery. Therefore, we propose the RObust Local EXplanations (ROLEX) method to develop robust, instance-level explanations for HPA models in this study. ROLEX adapts state-of-the-art methods and ameliorates their shortcomings in explaining individual-level predictions made by black-box machine learning models. Our analysis with a large real-world dataset related to a prevalent medical condition called fragility fracture and two publicly available healthcare datasets reveals that ROLEX outperforms widely accepted benchmark methods in terms of local faithfulness of explanations. In addition, ROLEX is more robust since it does not rely on extensive hyperparameter tuning or heuristic algorithms. Explanations generated by ROLEX, along with the prototype user interface presented in this study, have the potential to promote personalized care and precision medicine by providing patient-level interpretations and novel insights. We discuss the theoretical implications of our study in healthcare, big data, and design science. Collapse

Topics: explainable artificial intelligence usability user interface local interpretable model agnostic explanation problem solving

Methods: machine learning design artifact computational algorithm longitudinal research ablation analysis

Theories: lemon market theory

What Symptoms and How Long? An Interpretable AI Approach for Depression Detection in Social Media

2023 | International Conference on Information Systems | Citations: 0

Authors: Kuang, Junwei; Xie, Jiaheng; Yan, Zhijun

Abstract: Depression is the most prevalent and serious mental illness, which induces grav ... Expand

Abstract: Depression is the most prevalent and serious mental illness, which induces grave financial and societal ramifications. Depression detection is key for early intervention to mitigate those consequences. Such a high-stake decision inherently necessitates interpretability. Although a few depression detection studies attempt to explain the decision, these explanations misalign with the clinical depression diagnosis criterion that is based on depressive symptoms. To fill this gap, we develop a novel Multi-Scale Temporal Prototype Network (MSTPNet). MSTPNet innovatively detects and interprets depressive symptoms as well as how long they last. Extensive empirical analyses show that MSTPNet outperforms state-of-the-art depression detection methods. This result also reveals new symptoms that are unnoted in the survey approach. We further conduct a user study to demonstrate its superiority over the benchmarks in interpretability. This study contributes to IS literature with a novel interpretable deep learning model for depression detection in social media. Collapse

Topics: social media healthcare analytics

Methods: design artifact deep learning computational algorithm machine learning ablation analysis

Utilizing the omnipresent: Incorporating digital documents into predictive process monitoring using deep neural networks

2023 | Decision Support Systems | Citations: 0

Authors: Levich, Sergej; Lutz, Bernhard; Neumann, Dirk

Abstract: Predictive process monitoring (PPM) allows companies to improve the efficiency o ... Expand

Abstract: Predictive process monitoring (PPM) allows companies to improve the efficiency of their business processes by predicting aspects such as the process outcome, the next event, or the time until the next event. So far, existing studies have mainly focused on developing novel predictive models while using features solely from event logs. In this study, we aim to go beyond log data and increase the focus of PPM research towards external context information. To this end, we consider digital documents as they are omnipresent in many business processes and their inclusion can often be justified by a business rationale. However, incorporating digital documents into PPM models poses considerable challenges as they present unstructured data that can contain visual and textual cues of future process behavior, while manual feature extraction is generally not feasible. Therefore, we propose an approach that processes digital documents based on automated visual and textual feature extraction methods. Furthermore, we design a tailored integration module which transforms the extracted features from multiple document pages into a fixed-size representation that subsequently serves as input for the predictive models. Our evaluation, based on a real-world dataset of insurance claims from a mid-sized German insurance company, featuring 5131 process instances with 32,058 events and 39,242 document pages, shows that incorporating digital documents improves the performance by significant margins in predicting the damage type, the next event, and the time until the next event. Finally, we analyze how digital documents contribute to the model’s predictions in terms of Shapley additive explanations. Collapse

Topics: shapley additive explanation business process management artificial intelligence explainable artificial intelligence

Methods: BERT machine learning descriptive statistic artificial neural network business process modeling

A two-stage prediction model based on behavior mining in livestream e-commerce

2023 | Decision Support Systems | Citations: 0

Authors: Lin, Qinping; Jia, Ning; Chen, Liao; Zhong, Shiquan; Yang, Yuance; Gao, Tong

Abstract: Livestream e-commerce has been developing at a tremendous pace in recent years. ... Expand

Abstract: Livestream e-commerce has been developing at a tremendous pace in recent years. On livestream platforms, such as Douyin, a retailer attracts viewers into the live room through short video advertising, and then streamers promote and sell products in real time. In such a scenario, an accurate prediction of traffic and sales plays an essential role in operation management, including live content strategy and inventory control. However, complex behaviors (follow, share, comment, etc.) of users and long conversion paths (from seeing the advertisement to entering the live room, and to purchasing the goods) lead to poor performance of traditional prediction methods. Additionally, few studies focus on advertising information in evaluating live room performance. Therefore, we propose a two-stage learning model for traffic and sales prediction based on behavior mining, which combines marketing models and deep learning methods. In the first stage, we integrate user behaviors before getting into the live room with short video advertising data for traffic prediction. In the second stage, based on the traditional marketing model, AIDA (Attention-Interest-Desire-Action), we design a funnel convolutional neural network (FCNN) to learn sophisticated behaviors in the live room in both time and behavior orientations, and take the predicted traffic volume as the auxiliary information for sales prediction. Extensive experiments on real-world datasets from Douyin illustrate the efficacy of our proposed method, which shows the value of fusing marketing models with deep learning techniques. Furthermore, the in-depth analysis provides practical insights into user behaviors for livestream e-commerce. Collapse

Topics: electronic commerce advertising management user behavior marketing management personality

Methods: experiment deep learning time series analysis longitudinal research convolutional neural network

Theories: management theory